High-stakes #diplomacy is not only a political act. It is also a deeply psychological process that unfolds under stress, uncertainty, and prolonged exposure to hostile communication. This article examines how professional #counselling and structured psychological interventions can support diplomats, mediators, judges, and legal advisors who work inside long, tense, and often unresolved conflicts. Drawing on recent scholarship in #peace_psychology, negotiation research, judicial well-being studies, and the sociology of international institutions, the paper argues that psychological services are no longer a soft add-on to statecraft. They are becoming a core operational requirement. The paper reviews the main sources of #negotiation_fatigue, including cognitive overload, moral injury, secondary trauma, sleep debt, and identity-based threat. It then discusses evidence-informed interventions such as cognitive-behavioural coaching, mindfulness-based programmes, emotion regulation training, structured debriefing, restorative supervision, and trauma-informed judicial support. The paper also considers how counselling can be adapted to different institutional cultures, from foreign ministries and multilateral organisations to international courts and tribunals. It highlights the ethical dilemmas of confidentiality, national loyalty, and dual roles, and the risks of medicalising political disagreement. Finally, the paper suggests a research agenda focused on culturally responsive interventions, longitudinal studies of diplomatic careers, and integration of #mental_health services into training pipelines for negotiators and judicial officers. The paper is intended for students of international relations, law, and applied psychology who want to understand how the invisible emotional labour of #conflict_resolution can be systematically supported. Keywords: counselling, diplomacy, negotiation fatigue, conflict resolution, judicial well-being, mental health, international relations, mediation, moral injury, peace psychology. 1. Introduction Modern #international_relations rest on people who negotiate under conditions most citizens will never face. A senior diplomat may be asked to hold together a fragile ceasefire while sleeping four hours a night. A judge at an international tribunal may hear months of testimony on mass atrocities. A mediator may fly between capitals for a year without seeing a durable agreement. These are not ordinary jobs, and the psychological cost of doing them is now widely recognised in scholarly literature (Coleman, 2021; Halperin, 2024). Yet the institutional response to that cost has been slow and uneven. Many foreign ministries still treat #psychological_support as a private matter for the individual. Many courts still assume that legal training alone prepares officers for repeated exposure to graphic testimony. Both assumptions are increasingly hard to defend. This article takes a specific angle on that gap. It focuses on #counselling as an organised, professional practice, not as informal peer support or generic wellness programming. Counselling in this paper means a structured therapeutic relationship, or a structured intervention delivered by a trained mental health practitioner, aimed at helping the client manage stress, process difficult emotions, and make better decisions under pressure (Norcross and Lambert, 2019, in the tradition continued by Wampold and Owen, 2022). The clients considered here are not private citizens seeking help with personal issues. They are actors inside #high_stakes_diplomacy: diplomats, envoys, mediators, judges, prosecutors, defence counsel, court staff, and translators. Their working conditions are shaped by state interests, institutional hierarchies, and legal duties that private clients do not share. Counselling in this setting must therefore adjust to a very particular ecology. The paper has three main aims. First, to describe the psychological load of high-stakes diplomacy and judicial work in a way that is grounded in current research rather than in anecdote. Second, to review interventions that are supported by evidence, or at least by careful practice, and to explain how they can be adapted to diplomatic and judicial settings. Third, to identify the ethical, political, and cultural tensions that make this work harder than standard clinical practice. Throughout, the paper uses the term #negotiation_fatigue to describe the specific cluster of symptoms that appear when a person negotiates under stress for extended periods without adequate recovery. Negotiation fatigue is close to, but not identical with, burnout, moral injury, and secondary traumatic stress (Litz and Kerig, 2019, extended by Griffin and colleagues, 2022). It has features that are shaped by the negotiation task itself, including cognitive tunnel vision, decision fatigue, and a slow erosion of trust in the counterpart and in one's own judgment. The argument is organised around a simple claim. #Conflict_resolution at the international level is a human process before it is a legal or political one. If the humans doing that work are exhausted, morally injured, or emotionally dysregulated, the process suffers. Agreements are missed, tempers flare, positions harden, and legitimate compromises look like betrayal. Counselling, when done well, keeps the humans in the process functional. It does not replace strategy, law, or political will. It supports them. This claim is not new in occupational psychology, but it has only recently begun to enter mainstream discussions of statecraft and international law (Bar-Tal and colleagues, 2021; Shapiro, 2022). The paper is written for students. It uses plain language, short sentences where possible, and a structure that follows the conventions of #peer_reviewed journal articles. Section 2 reviews the literature on the psychology of diplomacy and judicial work. Section 3 sets out the theoretical framework, combining stress theory, moral injury theory, and social-psychological accounts of intergroup conflict. Section 4 describes the methodology used to select and evaluate interventions. Section 5 examines specific interventions for diplomats and mediators. Section 6 focuses on negotiation fatigue as a distinct clinical picture. Section 7 turns to judicial frameworks, including international criminal tribunals and human rights courts. Section 8 discusses ethical and cultural challenges. Section 9 sets out implications for training and institutional design. Section 10 acknowledges limitations. Section 11 concludes. Before moving on, one clarification is needed. This paper does not argue that counsellors should sit at the negotiation table or shape political outcomes. That would confuse professional roles and would rightly worry practitioners in both fields. The argument is more modest and, in a sense, more radical. It is that the state of mind of the negotiator, the mediator, and the judge is a legitimate object of professional care, and that this care is best delivered by trained #mental_health professionals working under clear ethical rules. When that care is missing, decisions get worse, careers get shorter, and institutions lose people they cannot easily replace. 2. Literature Review The literature relevant to this paper sits across four fields that do not always talk to each other. The first is peace and conflict psychology. The second is negotiation and mediation research. The third is judicial well-being and legal ethics. The fourth is clinical and counselling psychology, including the study of trauma, moral injury, and burnout in helping professions. Each field brings a piece of the picture. 2.1 Peace and conflict psychology Peace psychology has developed a rich account of how prolonged conflict changes the way people think, feel, and communicate. Recent work by Halperin (2024) and by Bar-Tal and colleagues (2021) shows how intergroup emotions such as anger, humiliation, and fear become stable features of protracted conflicts. These emotions are not just individual states. They are collective, socially reinforced, and passed on through education, media, and family narratives. When diplomats from societies locked in such conflicts sit across a table, they carry these emotions with them. Counselling that ignores this collective layer misses much of what is happening in the room. #Emotion_regulation therefore becomes a critical skill, not a soft one (Halperin, 2024). A related strand of work examines #reconciliation and its psychological preconditions. Studies of post-conflict societies suggest that reconciliation is rarely a single event and often takes generations (Nadler and Shnabel, 2020, whose framework continues to shape recent studies). For the diplomat, this has a practical implication. A negotiated settlement is often only the start of a long psychological process. Expecting rapid closure sets people up for disappointment and cynicism, both of which feed #negotiation_fatigue. 2.2 Negotiation and mediation research Negotiation research has traditionally focused on strategy, framing, and interests. Newer work has added the psychological cost of the process itself. Shapiro (2022) argues that identity threats are one of the least understood drivers of stalemate. When negotiators feel that their sense of self, or of their group, is under attack, they cannot reason clearly about substantive issues. Coleman (2021) develops a similar theme in his work on toxic polarisation, showing that negotiators inside deeply polarised systems face pressure not only from the counterpart but from their own constituencies. Ury (2024) has recently revisited these ideas in the context of what he calls the age of conflict, and he argues that inner work is now a legitimate part of the negotiator's preparation. Empirical studies of mediators, particularly in the family and community fields, have documented burnout and vicarious trauma at levels comparable to those found in emergency medical staff (McKibben and colleagues, 2021). International mediators face similar pressures, often amplified by long travel, isolation, and unclear reporting lines. Yet formal #supervision for international mediators is still rare (Lanz, 2021). 2.3 Judicial well-being and legal ethics The judicial well-being literature has grown quickly in the last five years. Studies of judges and court staff in national systems consistently find elevated rates of #secondary_traumatic_stress, sleep disturbance, and compassion fatigue, particularly among those working on cases involving violence, sexual abuse, or mass victimisation (Bourke, 2023; James, 2021). At the international level, judges and prosecutors at courts such as the International Criminal Court and the ad hoc tribunals have described similar patterns in reflective essays and interview studies (Sander, 2022). The literature also documents the ethical risk of unaddressed distress. Tired, morally injured, or emotionally dysregulated judges may make worse decisions and may erode public confidence in the process (Bourke, 2023). Legal ethics scholarship has begun to treat #well_being not as a personal virtue but as a professional duty. Bar associations in several jurisdictions have adopted policies that require or strongly encourage access to mental health services for lawyers and judges (Krill and colleagues, 2021). International courts have been slower, partly because they draw staff from many jurisdictions with different norms about #psychological_help. 2.4 Clinical and counselling psychology Finally, the clinical literature provides the tools that inform practice in these applied settings. The last decade has produced solid evidence for a small number of interventions that are directly relevant to high-stakes diplomacy. Cognitive-behavioural therapy remains the most extensively studied approach to anxiety, depression, and insomnia in working adults, including those in high-pressure roles (Hofmann and Hayes, 2019, whose framework is extended in more recent meta-analyses by Cuijpers and colleagues, 2023). #Mindfulness_based_stress_reduction and its variants have accumulated evidence for reducing perceived stress and improving emotion regulation in professional populations (Goyal and colleagues, 2022, updating earlier meta-analyses). Trauma-focused therapies, including cognitive processing therapy and eye movement desensitisation and reprocessing, have shown efficacy for post-traumatic stress in populations exposed to graphic material, including judges, jurors, and court staff (Hoge and colleagues, 2022). A further body of work examines the therapeutic relationship itself. The evidence remains robust that a strong working alliance, marked by agreement on goals, agreement on tasks, and a genuine emotional bond, predicts outcome across almost all therapeutic modalities (Wampold and Owen, 2022). In diplomatic and judicial contexts, where clients may be suspicious of any outsider and where confidentiality risks are unusually high, building this alliance takes longer and requires special attention to trust-building behaviours. Recent research on brief interventions has also grown, with structured single-session and two-to-four session models showing meaningful effects on stress, sleep, and emotional distress in working adults (Cuijpers and colleagues, 2023). These brief formats are particularly attractive in diplomatic settings, where longer treatment courses may be difficult to arrange around travel and operational demands. Two clinical concepts deserve special mention. The first is #moral_injury, which refers to the psychological harm caused by acts, or failures to act, that violate one's deep moral beliefs (Litz and Kerig, 2019; Griffin and colleagues, 2022). Diplomats and judges are exposed to moral injury when they are asked to accept, sign, or enforce outcomes they consider unjust. The second is #compassion_fatigue, which refers to the erosion of empathic capacity in professionals repeatedly exposed to the suffering of others (Figley and Ludick, 2022). Both concepts have moved from military and clinical settings into wider professional discussions, and both are highly relevant here. Taken together, these four literatures give a picture that is both worrying and hopeful. Worrying, because the psychological costs of high-stakes diplomacy and judicial work are real, measurable, and often ignored. Hopeful, because effective interventions exist and can be adapted to these settings if institutions are willing to invest in them. 3. Theoretical Framework This paper uses a three-layer framework to link the psychological experience of the individual diplomat or judge to the wider political process. The three layers are the individual, the interactional, and the institutional. Each layer draws on a different theoretical tradition, and each requires a different kind of intervention. 3.1 The individual layer: stress, moral injury, and identity At the individual layer, the framework draws on the transactional model of stress (Lazarus and Folkman, in the updated form used by contemporary researchers such as Crum and colleagues, 2020) and on moral injury theory (Litz and Kerig, 2019; Griffin and colleagues, 2022). Stress is understood as a function of the fit between perceived demands and perceived resources. When demands are high and resources are low, and when the person appraises the situation as threatening rather than challenging, chronic stress responses follow. In high-stakes #diplomacy, demands are often extreme and resources are often invisible. The negotiator may have a full political mandate but no one to talk to about the emotional impact of the work. Moral injury adds a moral dimension to this stress picture. A diplomat who signs a peace agreement that leaves victims without justice may experience a lasting sense of guilt or shame, even if the agreement was the best available. A judge who acquits because the evidence is thin may feel morally compromised, especially in atrocity cases. Traditional stress management does not address these moral wounds. They require a different kind of therapeutic work, one that acknowledges the reality of the moral conflict rather than trying to reframe it away (Griffin and colleagues, 2022). Identity is the third individual-level concept. Shapiro's (2022) work on #identity_based_conflict shows that when a person's core sense of self is threatened, cognitive flexibility drops sharply. Diplomats who represent their state carry not only their personal identity but also a professional and national one. Attacks on that identity, whether direct or symbolic, can trigger defensive responses that make agreement harder. 3.2 The interactional layer: emotion, communication, and trust At the interactional layer, the framework draws on social psychology of intergroup relations and on communication research. Halperin's (2024) work on emotions in conflict shows that shared emotional climates shape what is possible in a negotiation. Fear narrows the range of acceptable proposals. Anger makes concessions feel like weakness. Humiliation blocks any move that looks like accepting a lower status. Counselling at this layer aims to help the negotiator recognise these emotional currents in themselves and, cautiously, in the counterpart, without slipping into amateur psychology of the other side. #Trust is the other key interactional concept. Trust research suggests that trust builds slowly and breaks quickly, and that repair after a breach requires specific, credible acts, not just words (Kim and colleagues, 2023). Diplomats often work in environments where trust has been broken many times. Counselling can help the negotiator manage the emotional cost of continuing to engage a counterpart who has, in the past, acted in bad faith. The interactional layer also includes the dynamics inside the negotiator's own team. Delegations are small groups working under pressure, and they develop their own emotional climates. A delegation that has drifted into resentment, gossip, or scapegoating loses effectiveness even when its members are individually skilled (Coleman, 2021). Counsellors who work with delegations, or with heads of delegation, need to understand small-group dynamics as much as individual psychology. Facilitated team conversations, sometimes called reflective practice groups in clinical settings, can help delegations surface these dynamics before they harden. 3.3 The institutional layer: culture, hierarchy, and stigma At the institutional layer, the framework draws on organisational psychology and on studies of professional cultures. Foreign ministries, international organisations, and courts have distinctive cultures that shape how psychological distress is expressed, understood, and treated. Some cultures reward stoicism and treat any admission of distress as a career risk. Others have begun to normalise the use of counselling services, at least in some grades and functions (Krill and colleagues, 2021; Sander, 2022). #Stigma is a persistent institutional problem. Studies of legal and diplomatic populations consistently find that fear of career consequences is one of the strongest barriers to seeking help (Krill and colleagues, 2021). This finding shapes any serious proposal for counselling in these settings. Services must be structurally protected from career review processes, and their use must be normalised through leadership example and clear policy. The three layers interact. An individual with strong personal resources may still be overwhelmed by a toxic institutional culture. A supportive institution may still fail an individual whose interactional environment includes a hostile counterpart and a divided constituency. Effective counselling must therefore be part of a broader system, not a standalone service. 4. Methodology This paper is a conceptual and integrative review. It does not present new empirical data. It synthesises published research from 2021 onward, supplemented by a small number of earlier foundational works where they remain the reference points in their fields. The literature was selected through a purposive search of scholarly databases in psychology, law, international relations, and peace studies, using combinations of terms such as counselling, diplomacy, negotiation, mediation, judicial well-being, moral injury, and burnout. Preference was given to peer-reviewed articles, edited scholarly volumes, and reports from established research institutions. The paper follows a narrative review structure rather than a systematic one, because the field is too dispersed for a strict systematic review at present. Interventions were included if they were described in peer-reviewed sources, had at least some empirical support in comparable populations, and could plausibly be adapted to diplomatic or judicial settings. Interventions were excluded if they relied on techniques not compatible with basic ethical standards or if the evidence base was limited to a single small study. The paper is written for students, and technical detail has been kept at a level accessible to those with a general grounding in psychology, law, or international relations. Where clinical concepts are introduced, they are defined in plain terms. Where negotiation or legal concepts are used, they are illustrated with generic examples rather than with identifying case detail. No confidential material or identifying case information has been used. Two limitations of this approach should be noted here and returned to in Section 10. First, much of the strongest evidence on counselling interventions comes from populations other than diplomats and judges, such as military personnel, first responders, and healthcare workers. Generalisation to diplomatic and judicial populations is therefore reasoned rather than proven. Second, the paper draws mainly on English-language sources and on scholarship rooted in Western clinical traditions. Non-Western and non-English scholarship on these topics exists and is growing, but access to it in indexed databases is uneven. 5. Psychological Interventions in High-Stakes Diplomacy This section describes interventions that can be offered to diplomats and mediators. It groups them into four clusters: preparation, in-mission support, debriefing, and long-term care. 5.1 Preparation Preparation is the most neglected phase in most services. Diplomats are typically prepared politically and technically for a mission but rarely psychologically. A structured preparation programme can include three elements. The first is psychoeducation about the specific stressors of the mission, including sleep disruption, cultural adjustment, and the likely emotional profile of the counterpart's team. Psychoeducation is not therapy but it is protective, because it reduces the surprise factor when stress reactions appear (Cuijpers and colleagues, 2023). The second is skills training in #emotion_regulation, including brief techniques such as paced breathing, cognitive reappraisal, and grounding exercises that can be used at the table (Halperin, 2024). The third is a personal formulation exercise with a counsellor, in which the diplomat identifies their own vulnerabilities, such as trigger topics, past losses, or family situations that could interact with the mission. This formulation stays with the counsellor, not with the ministry, and it is used to shape support during the mission. 5.2 In-mission support During the mission, counselling can take several forms. Regular check-ins, typically brief and by secure telephone or video, allow the counsellor to monitor stress load, sleep, alcohol use, and interpersonal strain. These check-ins are not full therapy sessions. Their aim is early detection and simple support (Norcross and Lambert, 2019, in the framework updated by Wampold and Owen, 2022). When acute events occur, such as a collapse of talks, a hostile personal attack, or a traumatic incident on the ground, the counsellor can offer a focused intervention. Evidence for single-session psychological first aid is mixed, but structured brief interventions that combine grounding, meaning-making, and practical planning appear to be safe and often helpful in comparable populations (Hoge and colleagues, 2022). Formal debriefing sessions immediately after events, particularly those that require the person to relive the event in detail, are no longer recommended, because they can worsen post-traumatic symptoms (Hoge and colleagues, 2022). A special issue in-mission is #sleep. Chronic sleep restriction impairs judgment, empathy, and impulse control (Walker, 2021, whose synthesis remains influential in occupational settings). Counselling that ignores sleep is likely to fail. Simple cognitive-behavioural techniques for insomnia, adapted to the irregular schedule of diplomacy, can be taught in a few sessions and used during the mission. 5.3 Debriefing after major phases After a major phase of a negotiation, such as the closing of a round or the return from a long posting, structured debriefing with a counsellor can help the diplomat consolidate lessons and process emotions. Debriefing here is not the discredited single-session emotional discharge model. It is a set of short conversations over several weeks that help the person integrate the experience into their life story (Griffin and colleagues, 2022). Group debriefing with the negotiating team can also be useful, provided it is facilitated by a trained clinician and is separated from performance review. The purpose is to surface shared emotional patterns, such as team-level frustration or humiliation, and to prevent them from calcifying into resentments that will haunt the next round. 5.4 Long-term care Some diplomats will need longer-term care. This may be because of accumulated stress across many missions, personal life events, or specific mental health conditions that require treatment on their own terms. Long-term care may involve individual psychotherapy, medication, or both, and should be delivered by clinicians outside the immediate operational chain, with strict confidentiality. Long-term care is where questions of #confidentiality become most acute. In most jurisdictions, the same rules that apply to any patient apply to a diplomat, with narrow exceptions for imminent risk. Institutions that want to encourage use of long-term care must accept these rules and must not seek information from the counsellor about the content of sessions. Some services use contractual firewalls between the treating clinician and the employing ministry to make this explicit. 5.5 Reintegration after difficult postings Return from a difficult posting is a phase in its own right. Diplomats returning from conflict zones, long negotiations, or hostile capitals often report a sense of dislocation. Family and friends have moved on. Colleagues at headquarters may not understand what the returning diplomat has seen or done. The returning diplomat may feel simultaneously grateful to be safe and guilty about leaving colleagues, local staff, and interlocutors behind. Structured reintegration support, including a small number of counselling sessions in the first three months back, can help the returning person settle without minimising the experience. Peer groups of returned diplomats have also proved useful where they have been tried, offering a form of shared understanding that is hard to find elsewhere (Lanz, 2021). 5.6 Coaching and hybrid formats A growing area of practice is #executive_coaching for negotiators, sometimes delivered by clinicians and sometimes by non-clinical coaches with training in negotiation. Coaching can be a useful entry point for people who resist the label of therapy but who would benefit from structured reflection. It should not replace clinical care where clinical care is indicated. Institutions that offer coaching should be clear about the boundary between coaching and therapy and should ensure that clients can move between the two without stigma (Passmore, 2022). 6. Negotiation Fatigue as a Distinct Clinical Picture Negotiation fatigue is used in this paper as a working label for the specific pattern of psychological and cognitive changes that appear when a person negotiates under stress for extended periods. It overlaps with burnout, moral injury, and compassion fatigue, but it has features of its own. This section describes those features and outlines interventions. 6.1 Cognitive features Long negotiations produce a predictable set of cognitive changes. Attention narrows. The negotiator focuses on the last proposal, the last insult, or the last leaked story, and loses sight of the broader pattern. Decision fatigue sets in, and the negotiator increasingly relies on defaults, whether those defaults are their government's fallback position or their own habitual style. Creativity drops. Options that were considered early in the process are forgotten (Coleman, 2021; Ury, 2024). Counsellors working with fatigued negotiators can help them recognise these changes and use simple external supports. Structured decision journals, brief written summaries of each session written in a fixed format, and scheduled review points with a colleague or coach are low-cost tools that offset the cognitive narrowing. Sleep protection, as noted above, is central. 6.2 Emotional features Emotional features of negotiation fatigue include irritability, cynicism, emotional flatness, and a loss of curiosity about the counterpart. These features often appear together and can be mistaken for professional maturity. A senior diplomat who has stopped being curious about the other side is not experienced. They are tired. Coaching and short-term therapy can restore curiosity, in part by helping the person reconnect with the meaning of the work (Halperin, 2024). Emotional numbing is a warning sign that deserves particular attention. Numbing can be a defence against overwhelming affect, and in some cases it may indicate a post-traumatic response. When numbing appears, especially after specific traumatic events, referral for a full clinical assessment is appropriate. 6.3 Interpersonal features Fatigue affects relationships. The negotiator becomes shorter with staff, less patient with delegates, and more prone to interpret ambiguous behaviour as hostile. These changes affect not only the mission but also family and friends. Studies of diplomatic families are limited but consistently point to significant strain (Lanz, 2021). Counselling that ignores the family system misses a major source of both stress and support. Some services now include optional family-oriented sessions, delivered with strict rules about confidentiality and voluntary participation. 6.4 Physical features Physical features include sleep disturbance, appetite changes, gastrointestinal symptoms, headaches, and, in some cases, increased alcohol or other substance use. Substance use in high-pressure diplomatic environments is under-researched but clinically important. Screening tools such as brief alcohol use questionnaires can be included in routine check-ins, and any positive screen should trigger a supportive conversation, not a disciplinary one (Krill and colleagues, 2021). 6.5 Interventions specific to negotiation fatigue Interventions for negotiation fatigue combine general stress management with negotiation-specific techniques. General components include cognitive-behavioural work on unhelpful thoughts, mindfulness-based practice for emotion regulation, sleep hygiene and cognitive-behavioural therapy for insomnia, and physical activity as tolerated by the schedule (Goyal and colleagues, 2022; Cuijpers and colleagues, 2023). Negotiation-specific components include structured reflection on the process, scenario planning to reduce the load of real-time improvisation, and pre-agreed protocols for when to pause and consult (Ury, 2024). An important point is that #recovery is not simply the absence of demands. Real recovery requires active detachment from the work, which is difficult in an era of constant messaging. Institutions can support recovery by protecting off-hours, rotating personnel through demanding files, and making it clear that unavailability during agreed rest periods is not a career risk. Counsellors can support recovery by helping the client design realistic detachment routines. 7. Judicial Frameworks Judicial actors in international settings face a different but overlapping set of pressures. This section considers judges, prosecutors, defence counsel, and court staff at international courts and tribunals, and the transferable lessons for national judicial systems. 7.1 Exposure to graphic material International criminal courts and human rights bodies hear evidence about violence, torture, sexual violence, and mass atrocity. Exposure to this material, day after day, is a known risk factor for #secondary_traumatic_stress and vicarious traumatisation (Bourke, 2023; Sander, 2022). Interpreters and translators are particularly exposed, because they must render the material in first-person voice, and they often receive less institutional support than judges and lawyers (Sander, 2022). Interventions here draw on the trauma literature. Trauma-informed institutional practice includes limiting exposure to the most graphic material where possible, providing rotational relief, offering confidential clinical services, and training supervisors to recognise early signs of distress (Bourke, 2023). Trauma-focused therapies are effective for those who develop clinical-level symptoms (Hoge and colleagues, 2022). 7.2 Moral load Judicial actors face a distinctive moral load. Judges may be asked to apply legal rules that produce outcomes they personally find inadequate. Prosecutors may have to choose which cases to pursue and which to drop, knowing that some victims will not see justice. Defence counsel may be asked to defend clients accused of horrific acts. Each of these positions carries the risk of moral injury (Griffin and colleagues, 2022). Counselling for judicial moral injury requires care. The counsellor cannot resolve the moral tension by reframing. What they can do is help the person hold the tension without becoming disabled by it. This may involve values-based work, in which the person reconnects with the reasons they chose this career, and structured conversations about the limits of individual responsibility inside a larger system (Litz and Kerig, 2019). 7.3 Isolation and role constraints Judicial roles impose constraints on ordinary social support. A judge cannot discuss a pending case with friends or family. An international judge posted abroad may be far from their usual social network. These constraints make professional counselling particularly valuable, because it can be one of the few spaces where the person can talk openly about the emotional weight of the role (Bourke, 2023). Peer support, structured through formal programmes with trained peer counsellors, can complement professional services. Peer support has strong evidence in other high-responsibility professions and appears to be well received where it has been tried in judicial settings (James, 2021). 7.4 Institutional design Institutional design matters as much as individual services. Courts and tribunals that treat well-being as a core function, with dedicated staff, adequate budgets, and confidential pathways to care, produce better outcomes for their people than those that treat it as an afterthought (Sander, 2022; Krill and colleagues, 2021). Key design features include a clear separation between well-being services and performance management, senior leadership visibly using the services, and regular independent review of service quality. 7.5 Restorative supervision for judicial staff Restorative supervision is a model developed in health and social care that combines line management, professional development, and emotional support in a structured way (James, 2021). Applied to judicial settings, it means that court staff have regular sessions with a trained supervisor who is not their line manager and who can help them process the emotional demands of the work. Restorative supervision is not therapy, but it can prevent the drift toward clinical-level distress and can identify people who need referral to specialist services. Early pilot programmes in judicial contexts have reported positive reception, although formal evaluation is still limited (Bourke, 2023; James, 2021). 7.6 Transferable lessons for national systems National judicial systems can learn from the international experience. Judges and court staff in national systems also hear traumatic cases, also carry moral loads, and also face social constraints on ordinary support. Bar associations and judicial councils in several jurisdictions have begun to develop programmes drawing on the international experience (Krill and colleagues, 2021; James, 2021). These programmes emphasise confidential access, trained clinicians familiar with the legal environment, and normalisation of use through senior example. 8. Ethical, Cultural, and Political Challenges Counselling in high-stakes diplomacy and judicial settings faces ethical, cultural, and political challenges that go beyond ordinary clinical practice. This section addresses the most important of them. 8.1 Confidentiality and dual loyalty Confidentiality is the foundation of counselling. In diplomatic and judicial settings, it can come under pressure from state interests, security services, and management. Counsellors working in these settings need clear contractual and legal protection of confidentiality, with narrow, well-defined exceptions. Dual loyalty, in which the counsellor is asked to serve both the client and the employing institution, is a well-known ethical problem in occupational health and must be managed explicitly (Passmore, 2022). Written policies, independent supervision, and separation of clinical and management chains are the main safeguards. 8.2 Cultural responsiveness Diplomats and judges come from many cultures with different views on mental health, help-seeking, and emotional expression. Interventions designed in one cultural context cannot be simply exported. Culturally responsive counselling requires clinicians with training in cross-cultural work, awareness of their own cultural assumptions, and willingness to adapt techniques to the client's frame (Sue and colleagues, 2022, extending earlier foundational work). Language matters. Where possible, clients should have access to counsellors who speak their first language or a language they use fluently. 8.3 The risk of medicalising politics There is a real risk that framing political and moral disagreement as a mental health issue can serve as a way of dismissing legitimate positions. A diplomat who refuses to sign an unjust agreement is not necessarily suffering from moral injury in a clinical sense. They may be exercising sound moral judgment. Counsellors need to be alert to this risk and must not become instruments for making difficult but principled positions look pathological (Litz and Kerig, 2019; Griffin and colleagues, 2022). 8.4 Stigma and career risk As noted in Section 3, stigma is a persistent barrier to help-seeking in diplomatic and legal populations. Even where services exist, uptake is often low because of fear of career consequences. Reducing stigma requires more than communication campaigns. It requires structural changes, including protection of service records from career review, visible use by senior figures, and integration of well-being into ordinary performance conversations rather than treating it as a separate track (Krill and colleagues, 2021). 8.5 Security constraints Some counselling in these settings takes place under real security constraints. Sessions may need to be held in secure locations, over encrypted channels, or through cleared clinicians. These constraints add complexity but are manageable with appropriate planning. What is not acceptable is using security constraints as an excuse to deny access to services or to breach clinical confidentiality. 8.6 Boundaries with negotiation strategy A final concern is the boundary between counselling and negotiation strategy. Counsellors are not advisors on political tactics, and they should not be asked to profile the counterpart or to design manipulative strategies. Their role is to support the client's psychological functioning and, where relevant, to help the client use their emotional intelligence in the process. Confusing these roles damages both the counselling relationship and the negotiation. 9. Implications for Training and Institutional Design If the argument of this paper is accepted, several practical implications follow for how diplomats, mediators, and judicial officers are trained and supported. 9.1 Training pipelines Diplomatic academies, foreign service training programmes, and judicial colleges should include structured content on the psychology of #high_stakes_work. This content should go beyond a single wellness lecture. It should include practical skills in emotion regulation, communication under stress, recognition of stress and trauma responses in self and others, and use of professional support services. It should be delivered by clinicians with experience in these populations, in cooperation with senior practitioners who can lend credibility (Coleman, 2021; Ury, 2024). Training should also address the specific concept of #negotiation_fatigue as introduced here. Trainees should learn to recognise its features in themselves and colleagues, to use simple protective strategies, and to seek help without shame. 9.2 Career-long support systems Support cannot be limited to entry-level training. Diplomatic and judicial careers span decades, and the pattern of exposure changes over time. Institutions should build career-long support systems, with regular optional check-ins, easy access to counselling, and specific support around known transition points such as first missions to conflict zones, first exposure to atrocity evidence, and moves into senior leadership roles (Sander, 2022; James, 2021). 9.3 Team-level and organisational interventions Individual counselling is necessary but not sufficient. Team-level and organisational interventions matter as much or more. Team-level interventions include structured team debriefing, facilitated conversations about difficult ethical questions, and protected time for reflection. Organisational interventions include workload management, rotation policies, leadership training on well-being, and independent review of institutional culture (Passmore, 2022; Krill and colleagues, 2021). 9.4 Integration with legal and ethical frameworks Well-being services should be integrated with legal and ethical frameworks, not sit alongside them as an unrelated function. Codes of conduct for diplomats and judges should acknowledge the psychological demands of the work and the corresponding institutional duties. Ethics committees should have access to expertise on moral injury and burnout so that they can distinguish between misconduct and distress. 9.5 Research and evaluation Finally, services should be evaluated. The evidence base for counselling interventions in diplomatic and judicial populations is still thin, and every institution that offers services should contribute to building that base, subject to strict confidentiality protections. Areas of particular need include longitudinal studies of diplomatic and judicial careers, controlled evaluations of preparation and debriefing programmes, and cross-cultural studies of what works in which context. 10. Case Illustrations To make the argument concrete, this section offers three composite illustrations. None of them refers to any real individual or case. Each combines features common in the literature to illustrate typical challenges and responses. 10.1 The exhausted mediator A mediator has been shuttling between two capitals for eighteen months. Progress has been slow, and the process is now under public criticism. The mediator is sleeping poorly, drinking more in the evenings, and finding it hard to be patient with the delegations. A counsellor engaged through the mediation support unit works with the mediator on three fronts. First, sleep, using cognitive-behavioural techniques adapted to the irregular schedule. Second, cognitive framing of the criticism, distinguishing legitimate feedback from noise. Third, a values-based conversation about why the mediator took the assignment in the first place, which restores some of the lost meaning. The mediator does not become superhuman. They become functional again, which is enough to finish the round. 10.2 The judge after a long atrocity trial A judge has spent two years on a case involving mass violence, including graphic testimony from survivors. The verdict has been delivered. The judge feels flat, sleeps badly, and avoids news about the region. A structured post-case debriefing programme, spread over three months, gives the judge space to process the experience with a clinician trained in trauma work. Where post-traumatic symptoms are clinically significant, trauma-focused therapy is offered. The judge is also encouraged to take a genuine break before the next case. The programme is confidential, and its use is normalised by senior colleagues who have used it before. 10.3 The junior diplomat in a hostile posting A junior diplomat is posted to a hostile capital. The counterpart's officials are personally rude in meetings, and the diplomat's own ministry is providing little support. The diplomat begins to doubt their competence. A counsellor engaged remotely helps the diplomat separate personal identity from professional role, learn simple grounding techniques for use in meetings, and develop a plan for peer support with colleagues in similar postings. The counsellor also gently raises the question of whether the ministry's support arrangements should be reviewed, and helps the diplomat consider how to raise this through appropriate channels without damaging their career. These illustrations show that the interventions discussed in this paper are not exotic. They are ordinary clinical and coaching techniques, applied with sensitivity to the specific context of high-stakes diplomacy and judicial work. 11. Directions for Future Research The research agenda in this field is wide open. Several directions stand out. First, longitudinal studies of diplomatic and judicial careers, tracking stress, mental health, and career outcomes over many years, would give a much better picture of what accumulated exposure does and what protects against it. Such studies are difficult to run because of confidentiality, career mobility, and small populations, but they are not impossible. Partnerships between research institutions and professional associations offer a plausible way forward. Second, controlled evaluations of specific interventions are needed. Preparation programmes, in-mission support models, debriefing formats, and reintegration packages have all been described in the practice literature but not rigorously tested in these populations. Even modest randomised or matched-comparison studies would add substantial value. Third, cross-cultural research on help-seeking and on culturally responsive interventions is a priority. Diplomatic and judicial services are international by nature, and services that work in one cultural context may fail in another. Comparative studies across regions and institutions would help identify what is universal and what must be adapted (Sue and colleagues, 2022). Fourth, research on local staff, interpreters, and translators is urgently needed. These groups are often the most exposed and the least supported. Any serious agenda for well-being in international institutions must include them (Sander, 2022). Fifth, the interaction between political developments and psychological load deserves study. When public support for an agreement collapses, when a mediator is publicly attacked, or when a court's legitimacy is challenged, the psychological effects on those inside the process are significant but poorly documented (Coleman, 2021). 12. Limitations Several limitations of this paper should be acknowledged. First, the evidence base is uneven. Some interventions, such as cognitive-behavioural therapy for insomnia and trauma-focused therapies for post-traumatic stress, have strong general evidence but limited direct evaluation in diplomatic and judicial populations. Other interventions, such as coaching for negotiators, have grown faster than the research on them. Second, the paper is written in English and draws primarily on English-language scholarship. Important work in other languages and traditions has probably been missed. Non-Western perspectives on mental health, help-seeking, and reconciliation are particularly important in this field and deserve more space than a single paper can give. Third, the paper does not address in detail the situation of local staff in international missions and courts, who often carry heavy loads and receive less institutional protection than international staff. This is a significant gap in the wider field and should be a priority for future work. Fourth, the paper focuses on counselling and does not cover the full range of well-being interventions that institutions can offer. Physical health, financial support, family services, and reintegration programmes after difficult postings are all important and interact with counselling in ways that deserve dedicated study. Fifth, ethical dilemmas discussed in Section 8 are treated at a general level. Specific ethical guidance for counsellors working in diplomatic and judicial settings would require sustained work by professional bodies in psychology, law, and international affairs. 13. Conclusion High-stakes diplomacy and judicial work are demanding in ways that ordinary occupational categories do not fully capture. Diplomats, mediators, judges, prosecutors, defence counsel, and court staff carry sustained cognitive, emotional, moral, and physical loads. When these loads are not managed, the human costs are high and the institutional costs, in the form of poor decisions, lost personnel, and damaged legitimacy, are higher still. Counselling, understood as structured professional care delivered by trained clinicians, is one of the tools that can help. This paper has argued that #counselling in these settings must be adapted to the specific ecology of high-stakes work. It must respect confidentiality under difficult conditions, be culturally responsive, and avoid medicalising legitimate political and moral disagreement. It must be integrated with preparation, in-mission support, debriefing, and long-term care. It must address the distinct clinical picture of #negotiation_fatigue, with its cognitive, emotional, interpersonal, and physical features. In judicial settings, it must take account of exposure to traumatic material, moral load, and the social isolation of the role. Above all, it must be embedded in institutional cultures that treat well-being as a serious operational concern rather than a private matter. For students entering these fields, the practical implications are clear. #Emotion_regulation, self-awareness, and use of professional support are not weaknesses to hide. They are core professional skills. The best diplomats and judges are not those who feel nothing. They are those who feel, notice, and manage what they feel, and who know when to ask for help. Institutions that train and support such people will do better, in the long run, than those that pretend the work has no human cost. There is still much to learn. The research agenda outlined in Section 9 is only a starting point. But the direction is clear. The invisible emotional labour of #international_relations and judicial work is becoming visible, and once visible it can be supported, studied, and improved. That is a modest but real form of progress in a field where progress often looks out of reach. References Bar-Tal, D., Halperin, E., and Sharvit, K. (2021). The social psychology of intractable conflicts: Celebrating the legacy of Daniel Bar-Tal. Springer. Bourke, J. (2023). Judicial well-being and vicarious trauma: Emerging evidence and institutional responses. Journal of Judicial Administration, 32(2), 87-104. Coleman, P. T. (2021). The way out: How to overcome toxic polarization. Columbia University Press. Crum, A. J., Jamieson, J. P., and Akinola, M. (2020). Optimizing stress: An integrated intervention for regulating stress responses. Emotion, 20(1), 120-125. https://doi.org/10.1037/emo0000670 Cuijpers, P., Miguel, C., Harrer, M., Plessen, C. Y., Ciharova, M., Ebert, D., and Karyotaki, E. (2023). Cognitive behavior therapy vs. control conditions, other psychotherapies, pharmacotherapies and combined treatment for depression: A comprehensive meta-analysis. World Psychiatry, 22(1), 105-115. https://doi.org/10.1002/wps.21069 Figley, C. R., and Ludick, M. (2022). Secondary traumatization and compassion fatigue. In S. N. Gold (Ed.), APA handbook of trauma psychology (2nd ed., pp. 573-594). American Psychological Association. Goyal, M., Singh, S., Sibinga, E. M. S., Gould, N. F., Rowland-Seymour, A., Sharma, R., Berger, Z., Sleicher, D., Maron, D. D., Shihab, H. M., Ranasinghe, P. D., Linn, S., Saha, S., Bass, E. B., and Haythornthwaite, J. A. (2022). Meditation programs for psychological stress and well-being: Updated evidence and implications for practice. JAMA Internal Medicine, 182(6), 620-629. Griffin, B. J., Purcell, N., Burkman, K., Litz, B. T., Bryan, C. J., Schmitz, M., Villierme, C., Walsh, J., and Maguen, S. (2022). Moral injury: An integrative review. Journal of Traumatic Stress, 35(1), 34-45. https://doi.org/10.1002/jts.22738 Halperin, E. (2024). Emotions in conflict: Inhibitors and facilitators of peace making (2nd ed.). Routledge. Hofmann, S. G., and Hayes, S. C. (2019). The future of intervention science: Process-based therapy. Clinical Psychological Science, 7(1), 37-50. Hoge, C. W., Yehuda, R., Castro, C. A., McFarlane, A. C., Vermetten, E., Jetly, R., Koenen, K. C., Greenberg, N., Shalev, A. Y., Rauch, S. A. M., Marmar, C. R., and Rothbaum, B. O. (2022). Unintended consequences of changing the definition of posttraumatic stress disorder in DSM-5. JAMA Psychiatry, 79(6), 517-518. James, C. (2021). Towards trauma-informed legal practice: A review. Psychiatry, Psychology and Law, 28(2), 275-299. https://doi.org/10.1080/13218719.2020.1816850 Kim, P. H., Ferrin, D. L., and Dirks, K. T. (2023). Trust repair: An integrative framework and research agenda. Annual Review of Organizational Psychology and Organizational Behavior, 10, 371-395. Krill, P. R., Thomas, H. M., Kramer, M. F., Degeneffe, N., and Anker, J. J. (2021). Stressed, lonely, and overcommitted: Predictors of lawyer suicide risk. Healthcare, 9(12), 1626. https://doi.org/10.3390/healthcare9121626 Lanz, D. (2021). Local mediators and the emotional dimensions of peace mediation. International Negotiation, 26(3), 405-427. Litz, B. T., and Kerig, P. K. (2019). Introduction to the special issue on moral injury: Conceptual challenges, methodological issues, and clinical applications. Journal of Traumatic Stress, 32(3), 341-349. McKibben, W. B., Cook, R. M., and Fickling, M. J. (2021). Cognitive complexity and vicarious trauma among counselors: A quantitative investigation. Journal of Counseling and Development, 99(2), 195-205. Nadler, A., and Shnabel, N. (2020). The needs-based model of reconciliation: Recent advances and applications. In L. Tropp (Ed.), The Oxford handbook of intergroup conflict (updated online ed.). Oxford University Press. Norcross, J. C., and Lambert, M. J. (Eds.). (2019). Psychotherapy relationships that work: Volume 1, Evidence-based therapist contributions (3rd ed.). Oxford University Press. Passmore, J. (Ed.). (2022). The coaches' handbook: The complete practitioner guide for professional coaches. Routledge. Sander, B. (2022). International criminal justice and the mental health of its staff: Institutional responsibilities. Journal of International Criminal Justice, 20(3), 601-628. Shapiro, D. L. (2022). From identity conflict to identity connection: The role of relational identity in negotiation. Negotiation Journal, 38(1), 5-30. Sue, D. W., Sue, D., Neville, H. A., and Smith, L. (2022). Counseling the culturally diverse: Theory and practice (9th ed.). Wiley. Ury, W. (2024). Possible: How we survive (and thrive) in an age of conflict. Harper Business. Walker, M. (2021). Sleep, cognition and mental health: Emerging themes for a new decade. Nature Reviews Neuroscience, 22(9), 553-568. Wampold, B. E., and Owen, J. (2022). The great psychotherapy debate: The evidence for what makes psychotherapy work (3rd ed.). Routledge. #counselling #diplomacy #negotiation_fatigue #conflict_resolution #international_relations #judicial_frameworks #peace_psychology #mental_health #moral_injury #mediation #emotion_regulation #high_stakes_negotiation #judicial_well_being #trauma_informed_practice #psychological_interventions

The rapid expansion of #Global_Mental_Health as a field of research, policy, and practice has produced an uneven traffic of ideas, tools, and standards moving from wealthier countries into the #educational_sectors and #clinical_sectors of lower and middle income nations. This article uses #world_systems_theory, first developed by Immanuel Wallerstein, to explain that traffic as a structural pattern rather than a neutral transfer of scientific knowledge. Drawing on recent scholarship in #critical_psychiatry, medical anthropology, decolonial studies, and health policy, the paper argues that dominant Western #therapeutic_modalities such as cognitive behavioural therapy, biomedical psychiatric diagnosis, and short manualised interventions are circulated through training programmes, donor funded projects, textbooks, and digital platforms in ways that mirror the older economic core and periphery relationship. Universities in the #periphery adopt curricula written in the #core, clinicians are trained on standardised protocols developed elsewhere, and local healing systems are pushed to the margins of official mental health policy. The article maps the main channels of this movement, discusses concrete examples from South Asia, sub Saharan Africa, and Latin America, and reviews growing efforts at #decolonizing_mental_health through indigenous, community based, and pluralistic approaches. The conclusion is not that Western knowledge should be rejected, but that the current arrangement produces #epistemic_injustice, weakens local knowledge production, and often fails the very populations it claims to help. A more balanced global mental health would require symmetric exchange, funding autonomy in peripheral universities, and formal recognition of non Western healing traditions. Keywords: global mental health; world systems theory; core periphery; therapeutic exportation; decolonizing psychiatry; cultural competence; low and middle income countries; mental health policy; epistemic injustice; #cultural_psychiatry 1. Introduction Mental health has become a central concern of international health policy in the last two decades. Once a marginal subject in global health circles, it now appears in the Sustainable Development Goals, in World Bank reports, and in the strategic plans of large private foundations (Patel et al., 2018; Kola et al., 2021). The movement usually called #Global_Mental_Health, or GMH, has produced impressive achievements. It has drawn attention to the enormous #treatment_gap in poorer countries, mobilised funding for services, and pushed governments to include mental health in their basic health packages (Kohrt et al., 2018; Kola et al., 2021). Yet the same movement has attracted persistent criticism. Scholars argue that GMH exports a narrow set of Western diagnostic categories, therapies, and training standards to societies whose distress may not fit those categories, and whose #healing_traditions are older and often more embedded in daily life than any imported protocol (Fernando, 2017; Mills, 2018; Bemme & Kirmayer, 2020; Cosgrove et al., 2020). This article uses #world_systems_theory to make sense of that pattern. Developed by Immanuel Wallerstein and extended by later scholars, world systems theory divides the modern world into a #core of dominant economies, a #periphery of dependent ones, and a #semi_periphery of intermediate states (Wallerstein, 2004; Chase Dunn & Grimes, 1995). Value, labour, and raw materials flow from periphery to core, while manufactured goods, capital, and cultural products flow in the opposite direction. Applied to mental health, this framework directs attention to the direction of flow of ideas, curricula, assessment scales, therapy manuals, and even professional identities. It asks who produces the standards, who is trained in them, whose language they are written in, and whose interests they finally serve. The purpose of this paper is to describe those flows, connect them to the wider political economy of knowledge, and evaluate their consequences for students and clinicians in the developing world. The argument advances in five moves. First, world systems theory is introduced and adapted for mental health. Second, the field of GMH is outlined as a #core_project with a specific history. Third, the main channels through which Western therapeutic modalities are moved into peripheral #educational_sectors and #clinical_sectors are described. Fourth, concrete cases are examined. Fifth, resistance and alternative approaches are reviewed, including calls for #decolonial_psychology and community based practice. Two clarifications are important at the start. This paper does not claim that Western therapies never help people in the Global South. Cognitive behavioural therapy, interpersonal therapy, and antipsychotic medication have documented benefits in many contexts (Patel et al., 2018; Singla et al., 2021). The concern is with the terms of the exchange and the systematic marginalisation of local knowledge that often accompanies the transfer. The paper also does not romanticise every #indigenous_healing practice. Some are exploitative and some are harmful. The point is that decisions about which knowledge is valid should not be made only in the core. 2. Theoretical Framework: World Systems Theory and Mental Health Immanuel Wallerstein described the modern world system as a single capitalist economy that emerged in Europe from the sixteenth century and gradually incorporated the whole planet (Wallerstein, 2004). Within that system, states are placed unequally. #Core states specialise in high wage, high technology, capital intensive production. #Peripheral states specialise in low wage, low technology, labour intensive production, often of raw materials. #Semi_peripheral states have a mixed profile and can move up or down over time. The relationship between core and periphery is not accidental but structural. The wealth of one depends on the extraction of value from the other (Chase Dunn & Grimes, 1995; Robinson, 2011). Wallerstein and later scholars extended this analysis to knowledge production, education, and culture. Universities in the core produce the theories, the journals, the peer reviewers, and the citation metrics that shape what counts as legitimate research (Connell, 2019). Universities in the periphery consume that knowledge and often depend on core institutions for funding, degrees, and international recognition. The result is what Raewyn Connell has called #Northern_theory, a body of thought that presents itself as universal but is in fact grounded in a specific geography, language, and set of interests (Connell, 2007; Connell, 2019). Applied to mental health, world systems theory suggests three propositions. First, the categories used to describe #mental_distress are not neutral discoveries but products of a specific history located mainly in Europe and North America. The Diagnostic and Statistical Manual of Mental Disorders, the International Classification of Diseases, and the manualised therapies attached to them, were developed by professional communities embedded in wealthy states (Horwitz, 2021; Cosgrove et al., 2020). Second, the movement of these tools into the periphery is neither purely charitable nor purely commercial. It is shaped by donor priorities, professional ambitions, publishing economies, and geopolitical interests (Mills, 2018; Bemme & Kirmayer, 2020). Third, the transfer changes both the exporter and the importer, but unequally. Core institutions gain data, publications, and prestige. Peripheral institutions gain some services and training but also lose ground for the development of their own frameworks (Fernando, 2017; Kohrt et al., 2021). It is useful to distinguish this argument from a simple #cultural_relativism. The critique is not that every culture has its own equally valid psychiatry and that they cannot speak to each other. The critique is that the exchange as currently organised is asymmetric, and that this asymmetry has material effects on training, research, clinical practice, and patient outcomes. World systems theory sharpens the point by connecting these effects to a broader pattern of unequal exchange that runs through the world economy as a whole. 3. Methodology This article is a #narrative_review that combines conceptual analysis with a reading of empirical studies published mostly between 2019 and 2025, supplemented by earlier foundational works when needed. Sources were identified through searches of academic databases in psychiatry, public health, medical anthropology, and higher education studies, using combinations of terms such as global mental health, decolonising psychiatry, task sharing, cognitive behavioural therapy in low and middle income countries, world systems, and cultural adaptation. The review is not systematic in the PRISMA sense. Its aim is to build an interpretive argument, not to test a specific effect size. This is a legitimate method for #theory_building in social science and health policy (Greenhalgh et al., 2018). Three inclusion principles guided source selection. Recent peer reviewed work was preferred, especially where it presented new empirical data on training or service delivery in the Global South. Work from authors based in or closely connected to peripheral and semi peripheral countries was actively sought, to avoid a review that only repeats core voices about the periphery. Finally, when older sources were used, this was because they contain concepts that later work relies on. Wallerstein, Connell, and Kleinman are examples. Limitations should be stated openly. The analysis is shaped by the languages the author reads and by the visibility of certain journals in international databases. Work published only in Spanish, Portuguese, Arabic, Swahili, Hindi, or other national languages is under represented. The paper therefore probably underestimates the intellectual diversity of the periphery itself, which is a common problem in this literature. 4. Global Mental Health as a Core Project The contemporary GMH movement can be dated to the mid 2000s, when a series of articles in The Lancet argued that mental disorders were among the leading causes of disability worldwide and that most people in low and middle income countries received no treatment (Patel et al., 2018 review this history; see also Kola et al., 2021). The World Health Organization launched the Mental Health Gap Action Programme in 2008, offering a set of clinical guidelines for non specialist workers. Universities in North America and Europe established GMH divisions, master's degrees, and #research_networks. Large foundations, notably Grand Challenges Canada and the Wellcome Trust, funded implementation trials in Africa, Asia, and Latin America (Kohrt et al., 2018; Kola et al., 2021). Two features of this history matter for the present argument. First, the intellectual centre of gravity of GMH sits in a small number of institutions. Programmes at Harvard, King's College London, Johns Hopkins, Melbourne, and a handful of others produce most of the field's leadership, most of the widely cited protocols, and most of the reviewers who evaluate submitted grants and papers (Cosgrove et al., 2020; Bemme & Kirmayer, 2020). Second, the framing of the problem, that there is a huge #treatment_gap that must be closed by scaling up evidence based interventions, is itself a policy choice with theoretical assumptions. It assumes that the interventions that count as evidence based, and that were mostly tested in high income settings, can be adapted with minor modifications for very different social contexts (Fernando, 2017; Mills, 2018). Critics have questioned both assumptions. Suman Fernando and colleagues argue that GMH treats social suffering, poverty, displacement, and structural violence as if they were medical problems that could be solved with therapy or medication (Fernando, 2017; Fernando & Moodley, 2018). China Mills describes how the movement pathologises normal responses to injustice and displaces political questions into individual clinical ones (Mills, 2018). Lisa Cosgrove and colleagues raise concerns about the influence of pharmaceutical industry ties on GMH leaders and journals (Cosgrove et al., 2020). Doerte Bemme and Laurence Kirmayer, working from an anthropological angle, note that GMH has become an #assemblage of policies, professional networks, digital platforms, and metrics that produces a specific model of the person and the disorder even when it does not intend to (Bemme & Kirmayer, 2020). Placing GMH inside a world systems frame does not require accepting every point of these critiques. It requires recognising that the field is not a neutral scientific enterprise but a concrete social formation with a location, a history, and a set of interests. Whether one supports its main directions or not, it is a project centred in the core and directed at the periphery. It is also worth noting the specific way that #funding_flows shape the project. Large donors, both public and private, set the priorities that determine which questions get asked and which interventions get tested. A funder interested in scalable, measurable, short duration programmes will not usually support long term ethnographic work, community organising against poverty, or the training of traditional healers, even when local partners consider these more important. Over time this selection effect shapes the entire evidence base that later gets described as showing what works in low and middle income countries. What has been shown to work is, in significant part, what has been funded to be studied. This is not a conspiracy but a normal consequence of how research economies operate. It becomes a problem when the resulting evidence base is treated as if it captured the full range of possible interventions rather than a filtered subset of them (Cosgrove et al., 2020; Kola et al., 2021; Bhakuni & Abimbola, 2021). A further point concerns the language of #scale. GMH documents frequently emphasise the need to scale up effective interventions to close the treatment gap. Scaling is presented as a technical operation, but it is also a political one. It privileges interventions that can be delivered in short, standardised units across many settings, and it disadvantages practices that depend on long relationships, deep cultural knowledge, or context specific meaning. Traditional healing practices, family based interventions, and community rituals do not scale in the same way that a manualised therapy does. Their exclusion from official policy is not because they do not work, but because they do not fit the operational logic of large scale programmes (Fernando, 2017; Bemme & Kirmayer, 2020; Ventevogel et al., 2021). 5. Channels of Exportation into Educational Sectors Universities are the first place where #therapeutic_modalities from the core enter peripheral societies. This happens through several overlapping channels. 5.1 Curricula and textbooks Undergraduate and postgraduate courses in psychology, psychiatry, social work, and counselling in most Global South universities are built around textbooks originally written in the United States or the United Kingdom (Adams et al., 2019; Ratele, 2019). The dominant reference is usually the DSM or the ICD, with additional readings from a small canon of North American authors. In many programmes, non Western frameworks appear only in an introductory chapter on culture or in an elective module. Adams and colleagues describe this as a #hidden_curriculum in which students learn that the real content of psychology is Northern and that local knowledge is at best a decorative supplement (Adams et al., 2019; Ratele, 2019). The consequence for students is not only intellectual. Because international jobs, higher degrees abroad, and prestigious publications depend on fluency in the Northern canon, students who invest heavily in local traditions may find their careers narrower. This creates a strong personal incentive to align with the core, even for students who begin their studies committed to indigenous perspectives (Connell, 2019; Ratele, 2019). 5.2 Foreign trained faculty and returning scholars A second channel is the movement of people. Many faculty in psychiatry departments and psychology schools in the Global South have received doctoral or postdoctoral training in the United States, the United Kingdom, Canada, or Australia (Kohrt et al., 2018). When they return, they bring back not only knowledge but also professional networks, methodological preferences, and often ongoing collaborations with their former mentors. This is not necessarily a bad thing. But when a large majority of senior staff has been trained in the same small set of institutions, it produces a homogenisation of the discipline that is out of proportion to the diversity of the populations served (Connell, 2019; Ratele, 2019). 5.3 Short courses, workshops, and certification schemes Beyond formal degrees, an expanding market of short courses in evidence based psychotherapies, especially #Cognitive_Behavioural_Therapy, dialectical behaviour therapy, and eye movement desensitisation and reprocessing, delivers Western modalities directly to practitioners in the periphery (Singla et al., 2021; Kohrt et al., 2021). These courses are usually delivered in English by trainers from core countries and end with certification that has value on the international job market. Training packages developed by WHO, such as the mhGAP intervention guide, work in a similar way, offering standardised modules that can be delivered to primary care staff with limited local adaptation (Keynejad et al., 2021; Kola et al., 2021). Task sharing programmes, in which lay counsellors or community health workers deliver structured interventions under specialist supervision, have received particular attention. The evidence that they can reduce symptoms in some populations is real (Singla et al., 2021; Kohrt et al., 2021). What is often overlooked is that the intervention itself, its manuals, its outcome measures, and its research protocols usually originate in the core. The workers are local; the framework is not. 5.4 Publishing economies A fourth, less visible channel is the structure of academic publishing. High impact journals are concentrated in a few core countries, use English as the working language, and rely on editors and reviewers drawn largely from core institutions (Connell, 2019). Peripheral scholars who want to publish in these venues must adopt their conventions of writing, argument, and citation. Local journals often struggle to gain indexing in international databases, which lowers their visibility and their attractiveness to authors and readers. The result is a #citation_economy that steadily draws attention and effort towards core produced knowledge. Open access has been proposed as a solution but has produced new inequalities. Article processing charges of two to three thousand US dollars are unaffordable for most peripheral authors. Waivers exist but are inconsistent, and the mere fact that they must be requested marks the requester as a supplicant (Ross Hellauer, 2022 for related debates). #Epistemic_injustice, in Miranda Fricker's sense of a systematic disadvantage in being heard as a knower, is built into these arrangements (Fricker, 2007; Bhakuni & Abimbola, 2021). 5.5 Digital platforms and online training The COVID 19 pandemic accelerated a shift towards digital training and #digital_mental_health platforms. Universities and clinics in the Global South rapidly adopted online supervision, webinars, and app based interventions produced in the core (Torous et al., 2021). These tools have expanded access to training but have also deepened dependence on infrastructure and content controlled by core actors. The default language, the default clinical vignettes, and the default outcome measures embedded in the platforms shape practice in ways that are rarely questioned because they appear as neutral technical features. 6. Channels of Exportation into Clinical Sectors The educational channels feed directly into clinical practice, but there are also mechanisms that operate at the level of services. 6.1 Policy templates and guidelines WHO documents, Lancet Commission reports, and donor strategies provide policy templates that are widely adopted by ministries of health in the Global South (Patel et al., 2018; Kola et al., 2021). These templates typically recommend integrated mental health care in primary settings, task sharing with non specialist workers, and use of a defined set of essential medicines and manualised psychotherapies. Because ministries in resource constrained states rarely have the capacity to develop their own guidelines from scratch, they adopt the templates with minor edits. Over time, national policies across very different regions come to resemble each other closely, even when the underlying epidemiology, workforce, and cultural context are quite different (Fernando, 2017; Kohrt et al., 2021). 6.2 Diagnostic instruments and screening tools A second mechanism is the widespread use of screening and outcome instruments originally validated in high income populations. The Patient Health Questionnaire, the Generalized Anxiety Disorder scale, the Hamilton depression rating scale, and similar tools appear in projects across Africa, Asia, and Latin America (Kohrt et al., 2018). Translation and cultural adaptation efforts exist, but often reduce to linguistic translation without deeper #cultural_adaptation of items and cut offs (Kaiser et al., 2019; Kohrt et al., 2021). Local #idioms_of_distress, such as thinking too much in parts of West Africa, nervios in Latin America, or hikikomori related distress in East Asia, are studied by anthropologists but rarely integrated into routine assessment (Kaiser et al., 2019). The effect is subtle but important. When the instrument sets the definition of the outcome, interventions are judged by whether they move that instrument. Local outcomes valued by patients and families, such as ability to fulfil family obligations, restore social relationships, or perform ritual duties, are absent from the metrics that determine which programmes get scaled up. 6.3 Manualised psychotherapies The exportation of #Cognitive_Behavioural_Therapy is the most studied example of the transfer of clinical modalities. Trials in Pakistan, India, Uganda, Kenya, Zimbabwe, and elsewhere have adapted CBT and related therapies for depression, anxiety, and perinatal mental health, often delivered by lay counsellors (Singla et al., 2021; Chibanda et al., 2020). Some of these trials, such as the Friendship Bench in Zimbabwe, have become models cited across the world. The Friendship Bench uses trained grandmothers to deliver problem solving therapy on wooden benches outside health clinics, adapted for the Shona language and cultural context (Chibanda et al., 2020). The Friendship Bench is often praised, correctly, as an example of local innovation. It is worth noting, however, that it still sits inside a broadly cognitive behavioural framework and reports outcomes on standardised depression scales. It is a hybrid that shows what deep adaptation can achieve, but also shows how far the core framework reaches even into apparently indigenous programmes. Other imported therapies have received less thorough adaptation and simply operate as translated manuals (Kaiser et al., 2019; Kohrt et al., 2021). 6.4 Pharmaceuticals The clinical sector is also shaped by the flow of #pharmaceuticals. Antidepressants, antipsychotics, and anxiolytics developed and licensed in core countries dominate essential medicine lists in the periphery (Whitaker & Cosgrove, 2015; Cosgrove et al., 2020). Access to newer, more expensive drugs is uneven, but the overall pattern is a marketplace in which local companies produce generics of molecules discovered elsewhere, and prescribing patterns follow guidelines written elsewhere. Concerns about #overprescription of antidepressants, especially in primary care, that have been raised in high income countries are now emerging in middle income ones too (Horwitz, 2021; Kola et al., 2021). The influence of pharmaceutical marketing on prescribing culture is another aspect of this flow. Marketing budgets, continuing medical education events, and sponsored research shape the way that clinicians in peripheral countries think about diagnostic categories and their treatment. Even when direct to consumer advertising is prohibited, professional education channels can carry similar messages. This does not mean that clinicians are simply manipulated, but it does mean that the frame within which they exercise their judgement is partly built by commercial actors located in the core (Whitaker & Cosgrove, 2015; Cosgrove et al., 2020; Horwitz, 2021). 6.5 Humanitarian and post conflict interventions In humanitarian settings, the exportation of therapeutic modalities is especially visible. International non governmental organisations deploy trauma focused interventions in refugee camps, in post conflict regions, and after natural disasters (Ventevogel et al., 2021). These interventions are often based on models of post traumatic stress disorder developed in North America and Europe. Anthropologists have long argued that they may not match the meanings that affected communities themselves give to their suffering, and that they can distract from political and material demands such as safety, land, or reparation (Summerfield, 2019; Fernando, 2017). Recent work has tried to move humanitarian mental health towards more community based, culturally grounded models, with partial success (Ventevogel et al., 2021; Tol et al., 2021). 7. Case Illustrations Three brief illustrations show how the general pattern plays out in different regional settings. These are not exhaustive; each region contains countertrends and reform movements. The point is to give texture to the argument. 7.1 India India has a large and long established mental health profession, a substantial pharmaceutical industry, and a national mental health programme that predates the current GMH wave (Patel et al., 2018; Math et al., 2019). At the same time, most postgraduate psychiatry departments organise their teaching around DSM and ICD categories, and clinical psychology master's programmes rely heavily on textbooks and manuals from the United States and the United Kingdom (Math et al., 2019; Jain & Orr, 2016 for older context). Research funded through international collaborations has produced important adaptations of interventions such as the Healthy Activity Program, based on behavioural activation, for depression in primary care (Patel et al., 2018; Singla et al., 2021). Yet indigenous frameworks, including approaches drawing on #Ayurveda, yogic psychology, and community based traditions, have limited standing in the formal curriculum and in reimbursement structures (Math et al., 2019). Students who wish to work seriously with these frameworks often have to do so on their own initiative, outside the mainstream track. 7.2 Sub Saharan Africa Across sub Saharan Africa, the shortage of psychiatrists is severe, and #task_sharing programmes have expanded rapidly (Kohrt et al., 2018; Kola et al., 2021). Countries such as Uganda, Kenya, Ethiopia, Zimbabwe, South Africa, and Nigeria host large research collaborations with core universities. These have produced valuable services and important scientific work, including on the Friendship Bench, on the mhGAP intervention guide, and on stepped care models (Chibanda et al., 2020; Kola et al., 2021; Keynejad et al., 2021). At the same time, traditional and faith based healers, who are the first port of call for many people in distress, remain formally outside the health system in most countries. Recent research shows that partnerships between clinical services and healers are feasible and often welcomed, but they still depend on the initiative of individual researchers rather than on stable policy (Gureje et al., 2020; Kpobi & Swartz, 2019). South African scholars, including Kopano Ratele and colleagues, have argued forcefully for an African centred psychology that treats local knowledge as a primary source rather than a supplement (Ratele, 2019). 7.3 Latin America Latin America has strong traditions of #social_psychiatry, community mental health, and liberation psychology, associated with figures such as Ignacio Martin Baro in El Salvador and with reforms in Brazil, Argentina, and Chile. Brazil's psychiatric reform, which reorganised services around community based centres, is often cited as a model of contextually grounded change (Onocko Campos, 2019). Nevertheless, imported cognitive behavioural and pharmacological models continue to expand, and manualised therapies from the core circulate widely (Onocko Campos, 2019; Bemme & Kirmayer, 2020). Indigenous mental health, including work with the Mapuche in Chile, with Andean communities in Peru and Bolivia, and with Guarani groups in Paraguay and Brazil, receives limited state support, although scholarship in this area has grown (Pedersen, 2021). 8. Consequences for Peripheral Educational and Clinical Sectors The mechanisms described above produce several outcomes that deserve careful attention. 8.1 Intellectual dependence The first outcome is a form of #intellectual_dependence. Faculty and students in peripheral universities work with concepts, methods, and journals whose priorities are set elsewhere. This is not equivalent to intellectual failure, since much valuable work is produced under these conditions. But it does mean that the research agenda of a Nigerian, Colombian, or Vietnamese university is often shaped more by what core funders are willing to support than by what local communities identify as their most pressing questions (Connell, 2019; Ratele, 2019; Bhakuni & Abimbola, 2021). 8.2 Weakening of local knowledge systems A second outcome is the gradual weakening of local #knowledge_systems. When traditional healers are excluded from professional recognition, their apprenticeship structures shrink, their status declines, and the transmission of their knowledge is disrupted (Kpobi & Swartz, 2019; Gureje et al., 2020). Younger practitioners face a choice between an official career, which requires alignment with the imported model, and an unofficial one with limited security. Over generations this produces a real loss of #cognitive_diversity, in the same way that biological diversity is lost when a landscape is monocropped. 8.3 Mismatch between diagnosis and distress A third outcome is a persistent mismatch between the diagnoses used in clinics and the distress that people actually bring to them. Anthropological studies from many regions show that patients describe experiences that do not map neatly onto depression, generalised anxiety disorder, or PTSD (Kaiser et al., 2019; Summerfield, 2019). When clinicians nevertheless apply these labels, patients may feel misunderstood, or they may adopt the labels and reshape their own narratives to fit. Either way, the meaning of suffering is quietly rearranged. 8.4 Uneven benefits A fourth outcome is that the benefits of GMH programmes, though real, are unevenly distributed. Urban middle class populations often gain earlier and better access to therapies, especially digital ones, than rural and marginalised groups (Kola et al., 2021). Programmes that require literacy, smartphones, or English language skills can widen rather than narrow existing inequalities. Even task sharing programmes, which are designed to reach the poorest, tend to be concentrated in areas where research partnerships have been established, and to fade when project funding ends (Kohrt et al., 2021). 8.5 Effects on the workforce Finally, the transfer affects the mental health workforce itself. Ambitious clinicians and researchers in peripheral countries have strong incentives to migrate towards better paid positions in core institutions, contributing to a #brain_drain that further weakens local capacity (Connell, 2019; Kola et al., 2021). Those who remain often occupy positions that combine local service delivery with external accountability to distant funders, which can create ethical and practical tensions. 8.6 The problem of measurement A further outcome, closely tied to the previous ones, concerns #measurement. Grants and publications require quantifiable outcomes, usually collected with scales developed for other populations. Once these scales are in place, they define what counts as improvement, what counts as failure, and what counts as noise. Clinicians and researchers who work with populations whose distress is expressed in different vocabularies find themselves under pressure to translate that distress into the categories the scale can register. The consequence is a slow erosion of clinical listening. What the patient actually says matters less than what can be marked on a form. This is not unique to peripheral settings, but the imbalance is sharper there because the tools are more distant from the culture and language in which distress is originally expressed (Kaiser et al., 2019; Summerfield, 2019; Kohrt et al., 2021). 8.7 Effects on students Students in peripheral programmes experience the situation in specific ways that deserve attention. They typically encounter two mental health worlds. In the classroom they learn a body of knowledge with a foreign origin, framed in English, oriented towards international standards, and rewarded by international recognition. In their families, communities, and clinical placements they encounter a different world of #healing_practices, religious and spiritual explanations, family based decision making, and social suffering that does not fit neatly into DSM categories. Many students describe the experience as a form of intellectual doubling in which they learn to speak two languages that rarely meet. A few are able to translate skilfully between them. Many struggle silently. Some abandon one side entirely, either dismissing local knowledge as unscientific or dismissing academic psychology as irrelevant. Better curricula and supervision could reduce this cost, but only if the underlying arrangement is recognised (Adams et al., 2019; Ratele, 2019; Fernando & Moodley, 2018). 9. Resistance and Alternatives The picture is not entirely one sided. There is a growing body of work, some of it produced within GMH itself, that seeks to alter the current arrangement. 9.1 Decolonising mental health The most visible strand is the movement for #decolonizing_mental_health. This includes scholars and practitioners who argue for a systematic revaluation of #non_Western_epistemologies in psychology and psychiatry (Fernando & Moodley, 2018; Ratele, 2019; Adams et al., 2019). Their proposals range from curricular reform, in which local frameworks are treated as primary rather than supplementary, to structural changes in publishing and funding that would redirect resources to peripheral institutions (Connell, 2019; Bhakuni & Abimbola, 2021). 9.2 Community based and pluralistic care A second strand emphasises #community_based_care and formal cooperation with traditional and faith based healers. Studies in Ghana, Nigeria, and elsewhere show that structured collaboration between biomedical services and healers can improve outcomes and reduce human rights abuses in traditional settings, without erasing the traditions themselves (Gureje et al., 2020; Kpobi & Swartz, 2019). Pluralistic models accept that people move between systems and design services accordingly. 9.3 Cultural adaptation done seriously A third strand accepts that some imported interventions can help and focuses on doing #cultural_adaptation seriously rather than superficially. This means not only translating manuals but changing content, examples, therapeutic goals, and outcome measures in dialogue with local communities (Kaiser et al., 2019; Chibanda et al., 2020; Singla et al., 2021). The Friendship Bench, adapted versions of interpersonal therapy in Uganda, and locally reworked parenting interventions offer models. Serious adaptation is expensive and slow, which is one reason it is often skipped. 9.4 Rethinking evidence A fourth strand rethinks what counts as #evidence. Randomised trials remain important, but they are limited in what they can capture. Ethnographic studies, participatory action research, and mixed methods approaches produce different kinds of knowledge that are essential for understanding whether an intervention actually fits its setting (Bemme & Kirmayer, 2020; Whitley et al., 2020). Some GMH programmes have begun to embed these methods alongside quantitative measures. 9.5 Structural and political framings A fifth strand insists on framing mental health in structural terms. Poverty, inequality, gender based violence, climate change, and displacement are major drivers of #mental_distress (Lund et al., 2018; Patel et al., 2018; Compton & Shim, 2020). Any serious response must address these conditions, not only the symptoms they produce. This framing pushes the field beyond a purely clinical model and connects mental health to broader development, human rights, and social justice agendas. 9.6 South South collaboration A final strand emphasises #South_South_collaboration. Rather than treating each peripheral country as an isolated recipient of core knowledge, this approach builds networks among researchers, clinicians, and communities in the Global South (Kola et al., 2021; Bhakuni & Abimbola, 2021). Regional journals, joint training programmes, and shared research priorities can begin to shift the geography of knowledge production, even without dramatic changes in global funding structures. 10. Discussion Bringing these threads together, several general points emerge. First, the exportation of Western therapeutic modalities is not a scandal to be denounced and then dismissed, nor a technical fix to be implemented and celebrated. It is a structural feature of the contemporary world system in mental health. Like other transfers between core and periphery, it produces gains and losses that are unequally distributed and often invisible to those who benefit most. Second, students and clinicians in the developing world are not passive recipients. They negotiate, adapt, resist, and sometimes redirect the imported material in creative ways. The Friendship Bench, the Brazilian community mental health reform, and the African centred psychology movement all show what such agency can produce. World systems theory does not deny this agency; it locates it within the structural conditions that shape what is possible (Chase Dunn & Grimes, 1995; Connell, 2019). Third, the way forward is unlikely to be a rejection of Western knowledge in favour of a pure indigenous alternative. #Epistemic_pluralism, in which multiple frameworks are recognised and allowed to inform each other, is a more realistic aim (Fernando & Moodley, 2018; Kirmayer et al., 2018; Bemme & Kirmayer, 2020). What matters is the terms of exchange. Symmetric exchange requires investment in peripheral universities and journals, formal recognition of local healing systems, funding structures that do not force peripheral researchers to speak in the voice of the core, and training curricula that treat local traditions as primary sources of knowledge. Fourth, students who will work in the field, whether as clinicians, researchers, or policymakers, need conceptual tools to see the structures they are entering. A student in Karachi, Nairobi, or Lima who learns #Cognitive_Behavioural_Therapy without also learning about the political economy of its distribution will be less prepared than one who learns both. This paper is a small contribution to that dual education. Fifth, the mental health field is not unique. The exportation of educational and clinical models from core to periphery follows patterns visible in maternal health, in HIV programmes, in nutrition, and in health financing more broadly (Storeng & Mishra, 2014; Kentikelenis & Rochford, 2019; Robinson, 2011). The specific dynamics of mental health, in which categories of the self are at stake, sharpen the questions but do not create them. It is also worth thinking about what genuine reform might look like at the level of a single university department. A department that wanted to move away from pure dependence on core produced knowledge could take several concrete steps. It could require every course to include readings from authors based in the region, in translation if needed. It could develop its own case book, drawn from local clinical experience and reviewed by local practitioners, that supplements or replaces imported vignettes. It could allocate protected time for faculty to publish in national and regional journals, not only in international ones, and it could weight such publications properly in promotion decisions. It could build formal relationships with traditional healers, community organisations, and patient groups, and involve them in teaching and supervision. It could evaluate its graduates not only by their scores on international exams but by their capacity to work respectfully across knowledge systems. None of these steps requires a revolution in global policy. Each is available to a department that decides to act. Together they could over time change the intellectual character of the workforce a region produces (Adams et al., 2019; Ratele, 2019; Connell, 2019). At the level of #clinical_services, similar reforms are possible. A hospital or clinic that recognises the plural reality of its patients can begin to keep records that include the healing traditions patients have consulted, the explanatory models they hold, and the family arrangements around their care. It can offer training in culturally grounded assessment to its junior staff, and it can build referral relationships with traditional and faith based healers rather than treating them as competitors. Some of these practices already exist in isolated pockets around the world; the task is to move them from exceptions to normal expectations (Gureje et al., 2020; Kpobi & Swartz, 2019; Chibanda et al., 2020). Some counter arguments deserve honest engagement. It might be said that world systems theory is too coarse to describe such a complex field. That is fair. World systems theory works best as a first level of analysis, one that identifies the broad pattern; it must be supplemented by finer grained empirical work at country and community levels. It might also be said that focusing on core and periphery risks obscuring important variation within each category, including social class, gender, disability, and ethnicity. That is also fair. The goal is not to replace those analyses but to add to them. Finally, it might be said that critique risks paralysis: if all interventions are compromised by structural inequality, why do anything? The answer is that critique and action are compatible. Clinicians and educators must continue their work, and they can do it better if they are honest about the arrangements they operate within. 11. Conclusion The exportation of Western therapeutic modalities into the educational and clinical sectors of the developing world is best understood not as a neutral transfer of scientific tools but as one strand of a broader #world_system in which knowledge, like capital, moves unequally between core and periphery. This movement has produced measurable gains in access to some forms of care and has drawn welcome attention to mental health as a global concern. It has also weakened local knowledge systems, narrowed the meanings of distress, and reinforced dependence on core institutions for the definition of what counts as valid mental health science. A more balanced arrangement would preserve the useful contributions of Western frameworks while investing seriously in peripheral universities, indigenous healing traditions, community based services, and South South collaboration. This is not a call to shut the borders of knowledge. It is a call to open them in both directions. Students who read this article may be among those who will build such an arrangement in their own contexts. That will require both technical competence and structural awareness. The two are compatible, and both are necessary. The unresolved question that most affects the future direction of the field is whether the core institutions that currently dominate GMH will accept the loss of authority that a genuinely symmetric exchange would require. Recent shifts, including growing acknowledgement of #epistemic_injustice in global health, suggest that the intellectual case is being won (Bhakuni & Abimbola, 2021; Abimbola & Pai, 2020). Whether the material and institutional case follows depends on choices that will be made in the next decade by ministries, funders, universities, and students themselves. References Abimbola, S., & Pai, M. (2020). Will global health survive its decolonisation? The Lancet, 396(10263), 1627 to 1628. doi:10.1016/S0140-6736(20)32417-X Adams, G., Estrada Villalta, S., Sullivan, D., & Markus, H. R. (2019). The psychology of neoliberalism and the neoliberalism of psychology. Journal of Social Issues, 75(1), 189 to 216. doi:10.1111/josi.12305 Bemme, D., & Kirmayer, L. J. (2020). Global mental health: Interdisciplinary challenges for a field in motion. Transcultural Psychiatry, 57(1), 3 to 18. doi:10.1177/1363461519898035 Bhakuni, H., & Abimbola, S. (2021). Epistemic injustice in academic global health. The Lancet Global Health, 9(10), e1465 to e1470. doi:10.1016/S2214-109X(21)00301-6 Chase Dunn, C., & Grimes, P. (1995). World systems analysis. Annual Review of Sociology, 21, 387 to 417. Chibanda, D., Verhey, R., Munetsi, E., Cowan, F. M., & Lund, C. (2020). Using a theory driven approach to develop and evaluate a complex mental health intervention: The Friendship Bench in Zimbabwe. International Journal of Mental Health Systems, 14, 21. doi:10.1186/s13033-020-00354-x Compton, M. T., & Shim, R. S. (2020). Mental illness prevention and mental health promotion: When, who, and how. Psychiatric Services, 71(9), 981 to 983. doi:10.1176/appi.ps.201900374 Connell, R. (2007). Southern theory: The global dynamics of knowledge in social science. Polity. Connell, R. (2019). The good university: What universities actually do and why it is time for radical change. Zed Books. Cosgrove, L., Mills, C., Karter, J. M., Mehta, A., & Kalathil, J. (2020). A critical review of the Lancet Commission on global mental health and sustainable development: Time for a paradigm change. Critical Public Health, 30(5), 624 to 631. doi:10.1080/09581596.2019.1667488 Fernando, S. (2017). Institutional racism in psychiatry and clinical psychology: Race matters in mental health. Palgrave Macmillan. Fernando, S., & Moodley, R. (Eds.). (2018). Global psychologies: Mental health and the Global South. Palgrave Macmillan. Fricker, M. (2007). Epistemic injustice: Power and the ethics of knowing. Oxford University Press. Greenhalgh, T., Thorne, S., & Malterud, K. (2018). Time to challenge the spurious hierarchy of systematic over narrative reviews. European Journal of Clinical Investigation, 48(6), e12931. doi:10.1111/eci.12931 Gureje, O., Appiah Poku, J., Bello, T., Kola, L., Araya, R., Chisholm, D., Esan, O., Harris, B., Makanjuola, V., Othieno, C., Price, L., & Seedat, S. (2020). Effect of collaborative care between traditional and faith healers and primary health care workers on psychosis outcomes in Nigeria and Ghana (COSIMPO): A cluster randomised controlled trial. The Lancet, 396(10251), 612 to 622. doi:10.1016/S0140-6736(20)30634-6 Horwitz, A. V. (2021). DSM: A history of psychiatry's bible. Johns Hopkins University Press. Kaiser, B. N., Ticao, C., Anoje, C., Minto, J., Boglosa, J., & Kohrt, B. A. (2019). Adapting culturally appropriate mental health screening tools for use among conflict affected and other vulnerable adolescents in Nigeria. Global Mental Health, 6, e10. doi:10.1017/gmh.2019.8 Kentikelenis, A., & Rochford, C. (2019). Power asymmetries in global governance for health: A conceptual framework for analysing the political economy of a Sustainable Development Goals era. Globalization and Health, 15(Suppl 1), 70. doi:10.1186/s12992-019-0511-9 Keynejad, R., Spagnolo, J., & Thornicroft, G. (2021). WHO mental health gap action programme (mhGAP) intervention guide: Updated systematic review on evidence and impact. Evidence Based Mental Health, 24(3), 124 to 130. doi:10.1136/ebmental-2021-300254 Kirmayer, L. J., Gomez Carrillo, A., & Veissiere, S. (2018). Culture and depression in global mental health: An ecosocial approach to the phenomenology of psychiatric disorders. Social Science and Medicine, 183, 163 to 168. Kohrt, B. A., Asher, L., Bhardwaj, A., Fazel, M., Jordans, M. J. D., Mutamba, B. B., Nadkarni, A., Pedersen, G. A., Singla, D. R., & Patel, V. (2018). The role of communities in mental health care in low and middle income countries: A meta review of components and competencies. International Journal of Environmental Research and Public Health, 15(6), 1279. doi:10.3390/ijerph15061279 Kohrt, B. A., Turner, E. L., Rai, S., Bhardwaj, A., Sikkema, K. J., Adelekun, A., Dhakal, M., Luitel, N. P., Lund, C., Patel, V., & Jordans, M. J. D. (2021). Reducing mental illness stigma in health care settings: Proof of concept for a social contact intervention to address what matters most for primary care providers. Social Science and Medicine, 250, 112852. doi:10.1016/j.socscimed.2020.112852 Kola, L., Kohrt, B. A., Hanlon, C., Naslund, J. A., Sikander, S., Balaji, M., Benjet, C., Cheung, E. Y. L., Eaton, J., Gonsalves, P., Hailemariam, M., Luitel, N. P., Machado, D. B., Misganaw, E., Omigbodun, O., Roberts, T., Salisbury, T. T., Shidhaye, R., Sunkel, C., ... Patel, V. (2021). COVID 19 mental health impact and responses in low income and middle income countries: Reimagining global mental health. The Lancet Psychiatry, 8(6), 535 to 550. doi:10.1016/S2215-0366(21)00025-0 Kpobi, L., & Swartz, L. (2019). Indigenous and faith healing in Ghana: A brief examination of the formalising process and collaborative efforts with the biomedical health system. African Journal of Primary Health Care and Family Medicine, 11(1), 1 to 5. doi:10.4102/phcfm.v11i1.2035 Lund, C., Brooke Sumner, C., Baingana, F., Baron, E. C., Breuer, E., Chandra, P., Haushofer, J., Herrman, H., Jordans, M., Kieling, C., Medina Mora, M. E., Morgan, E., Omigbodun, O., Tol, W., Patel, V., & Saxena, S. (2018). Social determinants of mental disorders and the Sustainable Development Goals: A systematic review of reviews. The Lancet Psychiatry, 5(4), 357 to 369. doi:10.1016/S2215-0366(18)30060-9 Math, S. B., Gowda, G. S., Basavaraju, V., Manjunatha, N., Kumar, C. N., Enara, A., Gowda, M., & Thirthalli, J. (2019). The rights of persons with disability act, 2016: Challenges and opportunities. Indian Journal of Psychiatry, 61(Suppl 4), S809 to S815. doi:10.4103/psychiatry.IndianJPsychiatry_105_19 Mills, C. (2018). From critical psychiatry to critical global mental health. Transcultural Psychiatry, 55(3), 267 to 279. Onocko Campos, R. T. (2019). Psychiatric reform in Brazil: An overview from 1978 to 2019. Salud Colectiva, 15, e2065. Patel, V., Saxena, S., Lund, C., Thornicroft, G., Baingana, F., Bolton, P., Chisholm, D., Collins, P. Y., Cooper, J. L., Eaton, J., Herrman, H., Herzallah, M. M., Huang, Y., Jordans, M. J. D., Kleinman, A., Medina Mora, M. E., Morgan, E., Niaz, U., Omigbodun, O., ... UnUtzer, J. (2018). The Lancet Commission on global mental health and sustainable development. The Lancet, 392(10157), 1553 to 1598. doi:10.1016/S0140-6736(18)31612-X Pedersen, D. (2021). Political violence, ethnic conflict and contemporary wars: Broad implications for health and social well being. Social Science and Medicine, 274, 113800. Ratele, K. (2019). The world looks like this from here: Thoughts on African psychology. Wits University Press. Robinson, W. I. (2011). Globalization and the sociology of Immanuel Wallerstein: A critical appraisal. International Sociology, 26(6), 723 to 745. Singla, D. R., Kohrt, B. A., Murray, L. K., Anand, A., Chorpita, B. F., & Patel, V. (2021). Psychological treatments for the world: Lessons from low and middle income countries. Annual Review of Clinical Psychology, 17, 129 to 155. doi:10.1146/annurev-clinpsy-081219-124613 Storeng, K. T., & Mishra, A. (2014). Politics and practices of global health: Critical ethnographies of health systems. Global Public Health, 9(8), 858 to 864. Summerfield, D. (2019). A critique of seven assumptions behind psychological trauma programmes in war affected areas. Social Science and Medicine, 245, 112681. Tol, W. A., Ager, A., Bizouerne, C., Bryant, R., El Chammay, R., Colebunders, R., Garcia Moreno, C., Hamdani, S. U., James, L. E., Jansen, S. C. J., Leku, M. R., Likindikoki, S., Panter Brick, C., Pluess, M., Robinson, C., Ruttenberg, L., Savage, K., Welton Mitchell, C., Hall, B. J., ... van Ommeren, M. (2021). Improving mental health and psychosocial wellbeing in humanitarian settings: Reflections on research funded through R2HC. Conflict and Health, 15, 71. doi:10.1186/s13031-021-00397-y Torous, J., Jan Myrick, K., Rauseo Ricupero, N., & Firth, J. (2021). Digital mental health and COVID 19: Using technology today to accelerate the curve on access and quality tomorrow. JMIR Mental Health, 7(3), e18848. doi:10.2196/18848 Ventevogel, P., Ryan, G. K., Kahi, V., & Kane, J. C. (2021). Capturing the essential: Revising the mental health categories in UNHCR's refugee health information system. Intervention, 17(1), 13 to 22. Wallerstein, I. (2004). World systems analysis: An introduction. Duke University Press. Whitaker, R., & Cosgrove, L. (2015). Psychiatry under the influence: Institutional corruption, social injury, and prescriptions for reform. Palgrave Macmillan. Whitley, R., Sitter, K. C., Adamson, G., & Carmichael, V. (2020). Ethnography and critical mental health research. In S. Fernando & R. Moodley (Eds.), Global psychologies: Mental health and the Global South. Palgrave Macmillan. #Global_Mental_Health #World_Systems_Theory #Core_Periphery #Decolonizing_Psychiatry #Therapeutic_Exportation #Cultural_Adaptation #Epistemic_Injustice #Low_And_Middle_Income_Countries #Task_Sharing #Mental_Health_Policy #Indigenous_Healing #Critical_Psychiatry #Cognitive_Behavioural_Therapy #Community_Mental_Health #South_South_Collaboration

The rapid spread of generative language models has quietly reshaped how counselors, trainees, and supervisors work behind the scenes. When these tools are used inside a clinic or a training program without formal approval, oversight, or documentation, they form what is now called #Shadow_AI. This article examines the presence of shadow AI in #clinical_supervision, with a focus on three domains where the risk is highest: #case_conceptualization, #diagnostics, and #counselor_training. The paper draws on recent scholarship from mental health informatics, counseling ethics, and digital psychiatry to describe how unregulated use of these tools can shape clinical reasoning, distort diagnostic patterns, weaken supervisory relationships, and expose clients to privacy harms. It also considers how trainees develop professional identity when part of their thinking is outsourced to a system that is not accountable to any licensing body. The paper argues that the profession cannot rely on prohibition alone, because informal use is already widespread. Instead, a layered framework is proposed that combines institutional policy, supervisory dialogue, curricular reform, and ongoing evaluation. The goal is not to reject the technology but to bring it out of the shadows and into a form of practice that protects clients, supports trainees, and keeps clinical judgment where it belongs, in the hands of the counselor. Keywords: shadow AI, generative language models, clinical supervision, case conceptualization, diagnostics, counselor training, ethics, mental health technology. 1. Introduction Over the last three years, #generative_language_models have moved from research labs into the daily routines of many helping professionals. Counselors write progress notes with them. Trainees ask them to reframe their case formulations. Supervisors sometimes use them to prepare feedback for a supervisee. Very little of this activity is documented, disclosed, or reviewed. Most of it happens on personal devices, outside the electronic health record, and outside the awareness of the clinic director or the training program. This informal, undocumented, and often unauthorized use of #artificial_intelligence tools inside professional settings is now widely described as #Shadow_AI. The term borrows from the older concept of shadow information technology, which described employees using unapproved software to get work done. Shadow AI is different in one important way. The tools involved are not simple productivity applications. They are large statistical models that generate text that looks like clinical reasoning, that offer diagnostic impressions, that suggest interventions, and that shape how the user thinks about a client. When a #counselor uses one of these systems to think through a case, the tool is not just helping with paperwork. It is quietly participating in the clinical process. The counseling professions have long treated supervision as the space where clinical judgment is developed, refined, and audited. #Supervision is where a trainee learns to hold uncertainty, to consider alternative hypotheses, and to notice the influence of their own assumptions. If part of this reasoning is now being shaped by a language model that neither the trainee nor the supervisor fully understands, then supervision itself is under pressure. The supervisor may be reviewing a case formulation that was co-authored by an algorithm without knowing it. The trainee may be presenting ideas as their own that were, in practice, generated in seconds by a chatbot the night before. There is a growing body of literature on the promise and the danger of these systems in mental health contexts (Stade et al., 2024; De Choudhury et al., 2023). Much of that work focuses on client-facing chatbots, on companion apps, or on formal integrations into electronic records. The shadow layer, where clinicians and trainees quietly use general purpose models like consumer chatbots, has received less attention. Yet this is where much of the actual influence is happening today, because these tools are free or cheap, easy to reach, and require no institutional agreement. This article examines that shadow layer in three specific domains. The first is #case_conceptualization, where a counselor builds a working model of the client's difficulties. The second is #diagnostics, where the counselor tries to place the client's presentation within a framework such as the DSM or the ICD. The third is #counselor_training, where trainees and supervisors work together to grow clinical skill. In each of these domains, the paper asks a simple question. What can go wrong when a generative model is quietly involved, and what can the profession do about it? The paper is written in plain language for students and practitioners, but it draws on a peer reviewed evidence base. It does not argue that #language_models should be banned from clinical work. That position is unrealistic and probably harmful, because it drives use even further into the shadows. Instead, the paper argues for #transparency, for supervisory dialogue about tool use, and for a professional infrastructure that can support responsible integration. The article proceeds as follows. Section 2 defines shadow AI in the clinical context. Section 3 traces the arrival of generative models in mental health practice. Sections 4, 5, and 6 examine risks in case conceptualization, diagnostics, and counselor training in turn. Section 7 addresses ethical, legal, and regulatory concerns. Section 8 focuses on #confidentiality and data protection. Section 9 examines the supervisor's dilemma. Section 10 proposes a framework for responsible integration. Section 11 addresses implications for counselor education. Section 12 outlines future research needs. Section 13 concludes. 2. Defining Shadow AI in Clinical Contexts Shadow AI is not a single behavior. It is a family of practices that share three features. First, an #AI_tool is being used inside a professional context. Second, that use is not authorized, documented, or reviewed by the institution. Third, the output of the tool is entering clinical work, either directly, through a note or a decision, or indirectly, through the clinician's thinking (Coghlan et al., 2023; Blease et al., 2024). The most common form of shadow AI in counseling is the use of general purpose chatbots to help with paperwork. A counselor pastes a set of session notes into a chatbot and asks it to write a formal progress note in a preferred format. This is often described by clinicians as harmless time saving. It is not. The paste operation may transmit #protected_health_information across borders, into a system whose data handling practices are unknown to the clinician and often unknown even to the vendor's own employees. A second form is the use of language models for #clinical_reasoning support. A trainee is stuck on a case. They open a chatbot and describe the client in identifiable or near identifiable terms. They ask for possible diagnoses, for treatment suggestions, or for a case formulation. The response is polished, confident, and often plausible. The trainee then presents the resulting formulation in supervision without disclosing how it was produced. A third form is the use of language models by supervisors themselves. A supervisor may draft feedback for a trainee using a chatbot, or may generate teaching materials, or may ask the model to summarize a supervisee's video recording of a session. Each of these uses raises the same questions of confidentiality, accuracy, and accountability, but with the added weight that the supervisor is a gatekeeper of the profession. A fourth form, less visible, is the use of generative AI to prepare for licensure examinations, to write reflective journals, or to complete academic assignments during training. This use shapes what the trainee learns to think of as their own voice. Over months and years of drafts generated by a model, the trainee's clinical writing style, and eventually their clinical thinking style, may become entangled with the model's patterns. What unites these forms is the absence of institutional visibility. The clinic does not know. The training program does not know. The client does not know. The regulator does not know. Yet the tool is in the room, participating in decisions that affect vulnerable people. This invisibility is the defining feature of shadow AI, and it is what makes the phenomenon different from earlier debates about clinical software (Minerva and Giubilini, 2023). It is also important to distinguish shadow AI from what might be called sanctioned AI. A clinic may deploy an approved documentation assistant with a signed business associate agreement, with a data processing agreement, and with clinician training. That is not shadow use, even if it involves the same underlying model. The problem is not the model. The problem is the absence of institutional structure around its use. This distinction matters, because policy responses that target the technology in general will fail to address the practices that actually create risk. 3. The Rise of Generative Language Models in Mental Health Practice Understanding shadow AI in clinical work requires a brief look at how quickly these tools moved into everyday use. Before 2022, general purpose language models were mostly a research curiosity. They existed, but they were awkward to use, difficult to access, and often unreliable in ways that were immediately obvious. That changed with the release of consumer chatbots that combined a friendly interface with much larger underlying models (Stade et al., 2024). Within months, mental health professionals began experimenting. Some of this experimentation was formal. Researchers examined whether models could support #screening, whether they could draft #psychoeducational content, whether they could help with clinical documentation, and whether they could simulate therapeutic dialogue for training (De Choudhury et al., 2023; Ma et al., 2023). Much of the experimentation, however, was informal. Practitioners simply started using the tools, quietly, in their own workflow. Several forces have driven the informal spread. The first is #administrative_burden. Mental health clinicians in most systems spend a substantial portion of their working day on documentation. Any tool that promises to reduce that time has strong appeal, and a chatbot that turns a set of shorthand notes into a formal progress note in seconds is genuinely attractive (Torous et al., 2021). The second is the shortage of #supervision resources. Many trainees, especially in under resourced settings, have limited access to experienced supervisors and turn to language models as a substitute source of feedback and reflection (Haber et al., 2024). The third is the appearance of authority. These models produce fluent text that sounds like it comes from a knowledgeable source, and users often overestimate its reliability. The mental health field is also unusual in the depth of its documentation and the flexibility of its language. Unlike domains where the correct answer to a clinical question can be checked against an objective test, much of counseling work involves narrative reasoning, formulation, and interpretation. This is precisely the kind of work at which language models appear to excel, because they are trained to produce plausible narrative text. That surface plausibility is part of what makes shadow use so appealing, and part of what makes it so risky (Grodniewicz and Hohol, 2023). At the same time, the profession has been slow to develop guidance. Professional bodies have issued cautious statements about the use of AI in clinical settings, but these statements often focus on formal deployments and say little about informal or personal use by individual clinicians (American Counseling Association, 2023). Training programs have been even slower, in part because faculty themselves are uncertain about how these tools should be used in a syllabus, in a case seminar, or in a clinical placement. The result is a gap. The technology is present. The use is widespread. The guidance is thin. Into that gap, shadow AI has grown. The three sections that follow examine what that growth looks like in the three domains where clinical stakes are highest. 4. Risks in Case Conceptualization #Case_conceptualization is the process by which a counselor develops a working understanding of a client. It is where symptoms, history, culture, context, and relationship come together into a formulation that guides treatment. A strong conceptualization is not simply a list of problems. It is a hypothesis about how the client's difficulties are connected, what maintains them, and what might change them. Building such a hypothesis requires time, curiosity, tolerance for uncertainty, and iterative revision as new information emerges (Corey, 2024). When a trainee or a counselor uses a #generative_model to help build a conceptualization, several risks appear. The first is what can be called #premature_closure. Language models produce polished, coherent narratives quickly. When a trainee reads such a narrative, it can feel like understanding has been achieved, when in fact the trainee has simply received a plausible story. The temptation to accept this story and move on is strong, especially when the trainee is tired, stressed, or facing a deadline. Premature closure is a well documented cognitive risk in clinical work generally, and language models appear to intensify it (Salihu et al., 2024). A second risk is #anchoring. Once a model produces a formulation, the counselor's subsequent thinking tends to move around that formulation rather than developing independent alternatives. Even if the counselor is critical of the initial output, the shape of the model's proposal has already framed what feels relevant and what feels marginal. In case conceptualization, where the choice of framing can determine which pieces of client history are treated as central and which are ignored, this anchoring effect can quietly distort clinical judgment (Blease et al., 2024). A third risk concerns #cultural_context. Generative models are trained on vast text corpora that overrepresent English language material from a small number of countries. The formulations they produce tend to reflect assumptions from those contexts, including individualist framings of psychological difficulty, particular vocabularies of distress, and specific ideas about what a healthy self looks like (Coghlan et al., 2023). When a counselor uses such a model to conceptualize a client from a different cultural background, the resulting formulation may embed assumptions that do not fit. Because the output is fluent and confident, the mismatch is easy to miss. A fourth risk is #hallucination. Language models sometimes produce content that is fluent but false. In case conceptualization, this can mean citing a nonexistent construct, misattributing a theory, or asserting empirical claims that are not supported. A trainee who takes such content at face value may incorporate false premises into their formulation. Because the trainee often lacks the depth of literature familiarity to detect the error, and because the supervisor may not know that a model was used, the error can pass through several rounds of case discussion undetected (Minerva and Giubilini, 2023). A fifth risk is #erosion_of_clinical_curiosity. Case conceptualization is a skill that develops through struggle. The trainee learns by sitting with ambiguity, generating hypotheses, comparing them, revising them, and testing them against new information from the client. When a model provides an immediate synthesis, the struggle is removed. Over time, the trainee may lose the habit of tolerating uncertainty, which is one of the core professional capacities that supervision is meant to develop (Bernard and Goodyear, 2019). A sixth risk relates to the #therapeutic_relationship. Conceptualization is not a purely intellectual exercise. It is bound up with how the counselor understands and holds the client emotionally. A formulation that emerges from a chatbot, however clever, is not embedded in the counselor's felt sense of the client. When the counselor imports it into their thinking, there is a risk that the therapeutic relationship becomes organized around an idea the counselor has not lived through, rather than around the counselor's own developing understanding of the person in front of them. None of this means that language models cannot contribute to case conceptualization. They can play a useful role as a source of alternative hypotheses, as a check on the counselor's own thinking, or as a way to see how a case might look from a different theoretical orientation. But those uses only work when the counselor already has a strong independent formulation and uses the model as a challenge rather than a shortcut. In shadow use, the sequence is often reversed. The model produces the first draft, and the counselor's own thinking never fully forms. For supervisors, the practical implication is that questions about case conceptualization now need to include questions about process. It is no longer enough to ask what the formulation is. The supervisor also needs to ask how the trainee arrived at it. This can be done without accusation, framed as normal curiosity about the trainee's reasoning process, but it needs to be done routinely. 5. Risks in Diagnostics The diagnostic use of generative models raises a distinct set of concerns. Diagnosis in mental health is already contested. Categories change across editions of the DSM and ICD. Comorbidity is common. Presentations vary across cultures and life stages. Trained clinicians disagree, sometimes substantially, on the correct diagnostic label even when they interview the same client. Into this contested landscape, generative AI arrives with a confident manner and often plausible sounding output (Blease et al., 2024). The first diagnostic risk is #false_precision. When a clinician asks a language model for a differential diagnosis, the model typically produces a ranked list of possibilities with brief justifications. The list looks like the output of a careful clinician. In practice, it is a statistical continuation of the prompt. The model has not interviewed the client, has not observed nonverbal signals, has not taken a full history, and has no clinical accountability. Yet the format of the output invites the user to treat it as if these things had been done (Grodniewicz and Hohol, 2023). Studies of general medical use have found that #diagnostic_suggestions from language models can be both surprisingly good in narrow tasks and surprisingly wrong in ways that are hard to predict in advance (Stade et al., 2024). A second risk is the encoding of #diagnostic_bias. Language models learn from data that reflect existing patterns of clinical labeling. If those patterns include the historical overdiagnosis of certain conditions in some populations and the underdiagnosis of the same conditions in others, the model can reproduce and amplify those patterns. A clinician who uses the model as a diagnostic aid may therefore inherit biases that they would not have generated on their own, without being aware of the transfer (De Choudhury et al., 2023). A third risk is #symptom_matching without context. Language models are good at pattern matching between a described presentation and a set of criteria. They are less good at judging whether the described symptoms actually represent psychopathology in the client's context. A grieving client with intense low mood, a client from a culture where somatic idioms of distress predominate, or a young person going through a difficult developmental transition may all present with features that superficially match diagnostic criteria. A skilled clinician holds these presentations in context and weighs alternatives. A language model presented with a symptom list will often produce a diagnosis, because that is what the prompt asks it to do. A fourth risk concerns #differential_reasoning. Sound diagnostic practice involves not only naming what a client has but ruling out what they do not have. Language models can generate differentials, but their reasoning is often thin. They tend to list conditions rather than to explain the specific features that distinguish one from another in this particular client. A trainee who leans on the model may develop the habit of naming rather than reasoning, which is a deep problem in a field where accurate differential thinking is essential to safe care (Bernard and Goodyear, 2019). A fifth risk relates to #risk_assessment. Suicide risk, harm to others, and safeguarding concerns require careful, individualized judgment. Language models often produce generic risk statements and safety recommendations. These can be useful as reminders, but they are not substitutes for a clinician's assessment. In shadow use, there is a danger that a trainee will treat a model produced risk summary as sufficient documentation, when in fact the summary reflects nothing more than a probabilistic continuation of the prompt (Coghlan et al., 2023). A sixth risk is #diagnostic_inflation. Because language models are trained to be helpful and to respond to the specific prompt they receive, they can produce more elaborate diagnostic pictures than the case warrants. A brief description of low mood may return not only a depressive diagnosis but a series of ruled in secondary conditions. The user, particularly a trainee eager to demonstrate thoroughness, may accept the elaboration and enter it into the clinical record. Perhaps the most subtle diagnostic risk is #documentation_drift. Once a diagnosis, even a provisional one, appears in a note, it tends to persist. Subsequent clinicians read the note and treat the label as established. If the label originated in shadow AI use, that origin is invisible to everyone downstream. The clinical trajectory of a client may thus be shaped by a diagnostic impression that no human clinician ever independently formed, and that no supervisor ever reviewed. The counter argument, that language models could help with diagnostic accuracy by prompting clinicians to consider possibilities they might otherwise miss, has some merit. In sanctioned use, with careful evaluation and appropriate safeguards, such systems may add value. In shadow use, however, the safeguards are missing, and the plausible sounding output carries a weight it has not earned. 6. Risks in Counselor Training and Supervision The training of counselors is where the shadow AI problem becomes most acute, because training is where professional identity, judgment, and voice are formed. If trainees develop their clinical skills in constant collaboration with a language model, the skills that emerge are not the same skills that the profession has historically valued (Haber et al., 2024). A first concern is #skill_atrophy. Skills like case conceptualization, clinical writing, differential diagnosis, and reflective self examination are learned by doing them repeatedly, with feedback. If trainees outsource significant portions of that work to a model, the practice repetitions that build the skill never happen. The trainee produces work that looks competent, receives feedback that seems positive, and moves through the training program with a technically satisfactory record while their underlying skill development stalls (Salihu et al., 2024). A second concern is #evaluation_integrity. Training programs assess trainees through written case reports, reflective journals, treatment plans, and supervisory presentations. Each of these formats can be produced or heavily supported by a language model. A supervisor reading a well written case report cannot easily tell whether the report reflects the trainee's own reasoning or the reasoning of a chatbot with light edits. This undermines the evaluative function of the training program and, over time, undermines the credibility of licensure itself. A third concern is #reflective_practice. Reflective writing has long been a central tool in counselor training, because it develops the capacity to notice one's own reactions, biases, and blind spots. When trainees produce reflective writing through a chatbot, the writing may sound reflective without actually involving reflection. The trainee has not sat with the discomfort of noticing themselves. They have described that noticing to a machine, which has then produced a polished narrative on their behalf (Bernard and Goodyear, 2019). A fourth concern is #supervisory_alliance. Supervision depends on trust. The trainee must be willing to disclose uncertainty, mistakes, and confusion. If trainees use shadow AI without disclosure and worry about being caught, the honesty of supervision is compromised. Supervisors, sensing that something is off but unable to name it, may become suspicious in ways that damage the alliance. The result is a supervisory relationship that is quietly deformed by the presence of an undisclosed third party. A fifth concern is #identity_formation. Becoming a counselor involves more than acquiring techniques. It involves developing a voice, a set of values, a stance toward suffering, and a professional identity. This identity is built through years of practice, feedback, mistakes, and integration. When part of that process is delegated to a language model, the resulting identity may be less fully the trainee's own. This is a subtle harm, difficult to measure, but it matters for the profession's long term coherence (Haber et al., 2024). A sixth concern is #uneven_access. Language models are not equally available across all trainees. Those with paid subscriptions to more capable models, faster internet, and more time to experiment have access to more powerful tools. In an unregulated environment, this creates a hidden inequity in training outcomes. Trainees from better resourced backgrounds may appear to produce stronger work, not because they have stronger clinical thinking but because they have better tools that are quietly doing part of the work. A seventh concern is the effect on the #supervisor_role itself. Supervisors are also users of these tools. A supervisor who drafts feedback with a chatbot, who summarizes trainee session recordings with a model, or who generates teaching materials from a prompt is participating in the same shadow economy. This is not necessarily wrong, but it changes what supervision is. If both parties in the supervisory dyad are quietly using AI to prepare, the interaction becomes something new, and the profession has not yet named what that new thing is. An eighth concern relates to #trainee_wellbeing. Counselor training is emotionally demanding. Language models can feel like a supportive presence, always available, always patient, never judgmental. Some trainees may use them not only for academic tasks but as informal emotional regulation tools. This blurs the boundary between professional development and personal support in ways that may or may not be helpful. It also raises questions about whether trainees are receiving adequate human support, or whether the availability of a chatbot is masking a shortage of real supervision and peer connection (Torous et al., 2021). Taken together, these concerns suggest that the presence of shadow AI in training programs is not a minor efficiency issue. It is a structural change in how counselors are being formed. Programs that do not name and address this change are, in effect, allowing an unnamed influence to shape their graduates. 7. Ethical, Legal, and Regulatory Considerations The ethics of shadow AI in counseling cut across several established principles. #Informed_consent is a foundational commitment in counseling. Clients agree to a certain kind of relationship with a certain kind of professional. When a counselor uses a language model to help think about the client, the client has not consented to that involvement. In many jurisdictions, this is not merely an ethical concern but a legal one, because health data protection laws restrict the transfer of identifiable information to third party processors without appropriate agreements in place (Blease et al., 2024). A second ethical principle is #competence. Counselors are expected to practice within the boundaries of their training and to use tools appropriately. Using a general purpose chatbot to assist with clinical reasoning is not a validated intervention. The counselor cannot claim familiarity with the tool's error patterns, its data handling, or its underlying mechanism. If harm results from the tool's influence on the counselor's thinking, questions of competence will follow (Coghlan et al., 2023). A third principle is #nonmaleficence. The obligation to do no harm requires the counselor to be aware of the risks of their own practice. Shadow AI use often occurs without any risk analysis at all. The clinician has not thought carefully about what might go wrong. They have simply reached for the tool because it saves time. In ethical terms, this is a failure of due care, regardless of whether harm actually materializes in a given case. A fourth principle concerns #fidelity to the profession. Counselors work within a framework of shared standards that give the profession its public credibility. Widespread undisclosed use of language models erodes that framework. If the public came to believe that formulations, diagnoses, and treatment plans were being routinely produced by chatbots, the trust that makes counseling possible would be damaged (American Counseling Association, 2023). Legally, the landscape is uneven. Data protection frameworks such as the General Data Protection Regulation in Europe and the Health Insurance Portability and Accountability Act in the United States impose obligations on any transfer of health data to a third party. In principle, pasting a client narrative into a consumer chatbot without a business associate agreement or a data processing agreement is a breach of those obligations. In practice, enforcement is limited, because most such use is invisible to regulators. This invisibility does not make the use lawful. It only makes it undetected (Minerva and Giubilini, 2023). Regulatory bodies have begun to respond. Some licensing boards have issued cautionary statements. Some professional associations have started to update ethical codes. Some training programs have introduced explicit policies on the use of AI in coursework. But these responses are uneven, slow, and often reactive. In the meantime, the day to day practice of the profession is changing faster than the rules that govern it. There is also a growing conversation about #liability. If a counselor's decision, shaped by an unregulated language model, contributes to a harmful outcome, who is responsible? The counselor bears clinical responsibility, but the counselor may not fully understand the tool. The tool provider disclaims responsibility in the terms of service. The employer may not have known the tool was being used. The result is a diffusion of accountability that may leave harmed clients without meaningful redress (Haber et al., 2024). A related concern is #algorithmic_opacity. Language models are complex statistical systems whose behavior is not fully understood even by their developers. When they produce outputs that shape clinical decisions, neither the clinician nor the client can meaningfully audit those outputs. This creates a form of practice in which important judgments are influenced by processes that no one can explain, which sits poorly with counseling's traditional commitment to transparent, reflective practice. The ethical response cannot be limited to prohibition. Realistic ethics must acknowledge that the tools are here and will not disappear. What is needed is a set of practices that make shadow use less attractive, that create sanctioned pathways for beneficial use, and that keep the practitioner accountable for the decisions that emerge from their work, regardless of what tools helped shape them. 8. Confidentiality and Data Protection Of all the risks associated with shadow AI, the erosion of #confidentiality is perhaps the most concrete. Counseling depends on clients being able to speak about the most sensitive parts of their lives with the assurance that what they say will be protected. When a counselor pastes a client narrative into a chatbot, that assurance is quietly broken (Blease et al., 2024). The specifics vary by service. Some chatbot providers store user inputs for training purposes. Some retain data for extended periods. Some transfer data across national borders. Some allow employees or contractors to review conversations for quality assurance. Almost none of these providers were designed with clinical confidentiality in mind. Their terms of service typically warn against submitting sensitive personal information, but the warning is easy to miss and easy to rationalize away (Minerva and Giubilini, 2023). Even when a clinician tries to protect confidentiality by removing names and identifying details, the residual information can be surprisingly identifying. A description of a client's occupation, city, family structure, and presenting problem can uniquely identify that person to anyone with sufficient context. In smaller communities, such reidentification is even easier. The clinician who believes they have anonymized the input may still have transmitted information that could be traced back to an individual (Coghlan et al., 2023). There is also the matter of #metadata. Even if the content of a query is stripped of identifiers, the account information, the time of use, and the pattern of queries may reveal information about the clinician's caseload. A cluster of queries about a particular presentation at a particular time can indicate that a specific client is being discussed. These indirect leaks are difficult to anticipate and even more difficult to prevent (Stade et al., 2024). Another concern is #data_persistence. Once information has been transmitted to a third party system, it is very difficult to retrieve or delete. Even if the clinician later deletes the conversation from their own view, the data may persist in backups, logs, or training pipelines. In the event of a data breach at the provider, that information could become public. The client, who never consented to the transfer, may then experience harm that they cannot connect back to their counseling relationship. #Aggregate_risk is also worth considering. If many clinicians in a training program or a clinic are informally using the same chatbot, the aggregate volume of #sensitive_information flowing to a single provider can be substantial. Even without any individual breach, the provider is accumulating a large corpus of mental health related content that was collected without any meaningful consent framework. The long term uses of that corpus are unknown. The practical implications for counselors are significant. Effective protection requires that counselors either avoid entering identifiable client information into general purpose chatbots at all, or use only sanctioned tools with appropriate legal agreements in place. It also requires that clinics and training programs actively support these choices by providing access to sanctioned tools where they exist, so that clinicians are not forced to choose between confidentiality and the efficiency benefits of AI assistance. Clients themselves may need to be informed. Some ethicists have argued that if a clinician is using AI in any form that touches the client's information, the client has a right to know and to consent (Sedlakova and Trachsel, 2023). This position is not yet standard in professional codes, but the direction of ethical thinking is moving toward greater #disclosure. 9. The Supervisor's Dilemma Clinical supervisors face a particular kind of difficulty. They are responsible for the professional development of their supervisees, for the quality of care that supervisees provide, and, in many jurisdictions, for the licensure decisions that determine whether supervisees enter the profession. Shadow AI complicates all three of these responsibilities (Bernard and Goodyear, 2019). The first horn of the dilemma is #detection. Supervisors cannot reliably tell whether a case formulation, a treatment plan, or a reflective essay has been produced with substantial help from a language model. Detection tools exist, but they are imperfect and have significant false positive and false negative rates. A supervisor who tries to police AI use through detection alone is likely to end up in awkward, and sometimes unfair, confrontations with trainees. The second horn is #disclosure. Supervisors could ask trainees to disclose any use of AI in their work. This creates a norm of transparency. However, disclosure only works if trainees feel safe to be honest. In an environment where AI use is stigmatized or punished, trainees will hide it. The supervisor must therefore create a supervisory relationship in which disclosure is expected, normalized, and treated as an opportunity for learning rather than as evidence of misconduct. The third horn is #modeling. Supervisors themselves use these tools. If a supervisor asks a trainee not to use language models while quietly using them to draft feedback, the trainee will eventually notice, and the supervisor's authority will suffer. Supervisors must therefore be transparent about their own use, which means the supervisor must have thought carefully about what constitutes appropriate use for a supervisor. The fourth horn is #evaluation. If a trainee's work has been substantially shaped by a language model, the supervisor's evaluation of that work is unreliable as a measure of the trainee's skill. The supervisor may pass a trainee whose independent capacity to conceptualize cases is thin. Alternatively, the supervisor may become suspicious of all work and fail to appreciate genuine progress. Neither outcome serves the trainee or the profession. A useful framing is to treat AI use as a #supervision_topic rather than a #supervision_problem. The supervisor's task is not to eliminate AI use but to help the trainee develop a mature, reflective relationship with these tools. This means asking regularly about how the trainee prepared their work, what tools they used, what they got from those tools, and how they checked the outputs. Over time, the supervisor helps the trainee build a professional practice in which AI is a named, examined presence rather than a hidden one. This shift also changes what supervisors themselves need. They need training in how to supervise in the presence of AI. They need permission from their institutions to explore these questions openly. They need policies that clarify what disclosure is expected, what uses are acceptable, and what boundaries are firm. Without this infrastructure, individual supervisors are left to improvise, and improvisation across a profession produces inconsistency that itself becomes a problem. 10. Toward Responsible Integration: A Framework Given the depth and breadth of the risks discussed so far, a purely prohibitive response to shadow AI is unlikely to succeed. Prohibitions drive practices underground, and shadow AI already occupies that underground. What the profession needs is a #framework for #responsible_integration that combines institutional policy, supervisory practice, and individual reflection. The first element is #institutional_policy. Clinics and training programs should develop clear, written policies on the use of AI tools. These policies should distinguish between sanctioned tools with appropriate legal and technical protections and unsanctioned general purpose chatbots. They should specify what kinds of information may or may not be entered into which tools. They should require documentation of AI involvement in clinical decisions. They should provide sanctioned alternatives for the tasks where AI genuinely helps, such as drafting nonclinical correspondence, generating psychoeducational materials from approved sources, or supporting scheduling and administrative work (Stade et al., 2024). The second element is #supervisory_practice. Supervision should routinely include questions about tool use. These questions should be part of the normal rhythm of supervision, not reserved for moments of suspicion. Supervisors should model transparency about their own use. When a piece of work shows signs of heavy AI involvement, the response should be curiosity and dialogue, not accusation. The goal is to build a supervisory culture in which AI use is a topic, not a taboo (Bernard and Goodyear, 2019). The third element is #curricular_reform. Training programs should teach explicit skills for working with language models. These skills include understanding what the models can and cannot do, recognizing common failure modes, checking outputs against evidence, and reflecting on the effect of tool use on one's own thinking. Programs should also teach the ethics of AI use, drawing on emerging scholarship in the field (Haber et al., 2024). Assessment should be redesigned to include tasks that cannot be completed without genuine independent thinking, such as live case discussions, oral examinations, and observed clinical work. The fourth element is #individual_reflection. Every counselor, whether a trainee or a licensed practitioner, should develop a personal practice of thinking about their own AI use. This practice should include noticing when they reach for a tool, why they reach for it, and what effect the tool has on their thinking. Over time, this reflective practice becomes part of the counselor's professional identity, in the same way that ongoing reflection on countertransference or on personal biases is now expected. The fifth element is #client_engagement. Where appropriate, clients should be informed about the counselor's use of AI. This does not require a technical briefing. It can be a simple statement that certain administrative tasks may be supported by technology, with an assurance about how their information is protected. Where AI is used in ways that more directly touch clinical decisions, more detailed disclosure and consent may be warranted (Sedlakova and Trachsel, 2023). The sixth element is #ongoing_evaluation. The field needs mechanisms to evaluate the effects of AI integration over time. This includes research on clinical outcomes, on training outcomes, on the quality of supervision, on client trust, and on professional identity. Without such evaluation, the profession will be shaped by these tools without knowing what shape it is taking. The seventh element is #community_of_practice. Individual counselors and small clinics cannot solve these problems alone. Professional associations, regulatory bodies, and academic institutions should convene ongoing conversations about AI in counseling. These conversations should include voices from technology, ethics, law, and clinical practice. They should also include client voices, because the ultimate purpose of the profession is client welfare. Finally, the framework should be #adaptive. The technology is changing quickly. Policies that seem sensible today may need revision within a year. A responsible framework builds in review cycles, welcomes feedback from practitioners on the ground, and treats itself as a work in progress rather than a finished document (Torous et al., 2021). The strength of this framework is that it does not depend on a single dramatic intervention. It depends on many small, sustained practices, each of which is achievable in ordinary institutional life. The weakness is that it requires resources and attention. Overworked clinics and understaffed training programs may find it difficult to invest in the necessary infrastructure. That is why the framework needs to be championed at multiple levels, not left to individual clinicians alone. 11. Implications for Counselor Education Counselor education is where the most urgent changes need to happen, because the current generation of trainees is the first to move through their formation with #generative_AI as an available companion. Programs that ignore this reality will graduate counselors whose relationship with these tools has been shaped entirely by informal experimentation. The first implication is that AI literacy needs to be treated as a core competency. Every counseling trainee should complete coursework that covers how language models work at a basic conceptual level, what their known failure modes are, how they interact with data protection frameworks, and how they can influence clinical thinking. This coursework should be integrated into ethics, into research methods, and into clinical skill training, not offered as a separate optional module (Haber et al., 2024). The second implication is that assessment design needs to change. Traditional take home assignments, in which a trainee writes a case report over the course of a week, are increasingly difficult to evaluate as measures of individual skill. Programs should complement these assignments with live case discussions, oral examinations, and observed clinical work. The goal is not to catch AI use but to ensure that graduation from the program certifies a genuine capacity to think clinically. The third implication concerns supervision of supervisors. Faculty and clinical supervisors themselves need support in developing the skills to supervise in an AI aware way. This support can take the form of workshops, communities of practice, and updated supervision manuals. It should not be left to individual supervisors to figure out on their own how to have these conversations with their supervisees. The fourth implication is that programs should provide #sanctioned_tools. If trainees have access to a program supported, ethically vetted AI environment for appropriate tasks, they are less likely to resort to unsanctioned consumer chatbots. This requires investment, but it is a form of investment that pays off in both risk reduction and educational quality. The fifth implication is that programs need to address equity. Not all trainees have equal access to AI tools. Programs should ensure that any expected use of AI is supported by institutionally provided access, so that ability to complete assignments does not depend on ability to pay for premium services (Torous et al., 2021). The sixth implication is that programs should engage in ongoing #research on their own graduates. What do counselors trained in an AI rich environment look like five and ten years after graduation? What skills do they have? What skills have they lost? What are their patterns of practice? These questions cannot be answered without longitudinal work, and counselor education programs are well positioned to support such work. Taken together, these implications suggest that counselor education is entering a period of significant change. Programs that engage this change thoughtfully will produce graduates who are equipped to work in an AI shaped clinical world. Programs that avoid it will produce graduates whose formation has been shaped by tools they were never taught to think about critically. 12. Future Research Directions The academic literature on AI in mental health has grown quickly, but shadow AI as a specific phenomenon has received limited empirical attention. Several research directions deserve priority. First, #prevalence studies are needed. How widespread is the informal use of language models among counselors and trainees? What tasks are involved? What are the patterns of use across settings and career stages? Reliable prevalence data would help policy responses match the reality on the ground rather than assumptions about it (De Choudhury et al., 2023). Second, #outcome_research is needed. Does shadow AI use affect the quality of case conceptualization, the accuracy of diagnosis, or the trajectory of therapy? Rigorous studies comparing AI assisted and unassisted clinical work in controlled conditions would help ground the current debate in evidence. Third, #training_outcomes research is needed. How does AI use during training affect the eventual clinical capacity of graduates? Longitudinal studies of counselor education programs with different AI policies could shed light on this question. Fourth, ethical and legal analysis needs to continue. As courts, regulators, and professional bodies respond to AI use, the resulting body of guidance needs to be examined, compared across jurisdictions, and refined (Blease et al., 2024). Fifth, #client_perspective research is essential. What do clients know about their counselors' AI use? What do they think about it? What do they want to know? What kind of disclosure feels respectful and useful? These questions can only be answered by asking clients directly (Sedlakova and Trachsel, 2023). Sixth, comparative studies of sanctioned and shadow use would be valuable. If a clinic implements a sanctioned AI system, does shadow use decrease? What conditions support the shift? Seventh, research on supervisor experience is needed. How do supervisors currently manage AI in supervision? What has worked? What has not? Practice based evidence from experienced supervisors could inform guidance for the next generation (Bernard and Goodyear, 2019). Finally, the field needs methodological innovation. Traditional research methods, which assume that the technology under study is stable, are strained by the pace of AI development. Researchers need new approaches that can produce useful findings at a speed that matches the speed of technological change, without sacrificing rigor. 13. Conclusion Shadow AI is already in the room. It is in the clinician's late night documentation session, in the trainee's rushed case report, in the supervisor's feedback drafting, and in the countless small workflow shortcuts that shape the day to day of clinical practice. Pretending otherwise does not make it go away. It only ensures that the profession's response arrives too late and applies to a practice that has already changed. The risks are real. Case conceptualization can be quietly distorted by anchoring, premature closure, and cultural mismatch. Diagnostics can be shaped by false precision, encoded bias, and documentation drift. Counselor training can be undermined by skill atrophy, evaluation problems, and quiet erosion of the supervisory alliance. Client confidentiality can be broken by pastes into consumer chatbots that were never designed for clinical use. These are not speculative concerns. They are visible in current practice and are documented in a growing literature. At the same time, a response built solely on prohibition will fail. The tools are too accessible, too useful, and too integrated into the digital environment in which clinicians now work. The realistic response is to bring these tools out of the shadows through clear policy, active supervisory dialogue, thoughtful curricular reform, and honest engagement with clients. The goal is not to purge AI from counseling. The goal is to keep clinical judgment where it belongs, in the hands of a reflective, well trained, well supported counselor who knows what tools they are using, why they are using them, and how those tools affect their thinking. For students entering the profession now, this is not an abstract debate. It is the environment in which they will practice for the rest of their careers. Building a personal habit of transparency about tool use, of critical evaluation of AI outputs, of ongoing reflection on the effect of these tools on one's own clinical mind, is one of the most important professional habits a new counselor can develop. Programs, supervisors, clinics, regulators, and professional associations all have their part to play, but the individual counselor's habits of thought matter as much as any policy. The next several years will decide whether the profession shapes its relationship with generative AI or whether that relationship shapes the profession. The evidence and the arguments reviewed in this article suggest that the window for making that choice deliberately is still open, but it is closing. What is needed now is not more warnings, but sustained, practical work at every level of the field. The clients whom counselors serve deserve nothing less. References American Counseling Association. (2023). ACA guidelines for the use of artificial intelligence in counseling practice. American Counseling Association. Bernard, J. M., and Goodyear, R. K. (2019). Fundamentals of clinical supervision (6th ed.). Pearson. Blease, C., Worthen, A., and Torous, J. (2024). Psychiatrists experiences and opinions of generative artificial intelligence in mental healthcare: An online mixed methods survey. Psychiatry Research, 333, 115724. https://doi.org/10.1016/j.psychres.2024.115724 Coghlan, S., Leins, K., Sheldrick, S., Cheong, M., Gooding, P., and D Alfonso, S. (2023). To chat or bot to chat: Ethical issues with using chatbots in mental health. Digital Health, 9. https://doi.org/10.1177/20552076231183542 Corey, G. (2024). Theory and practice of counseling and psychotherapy (11th ed.). Cengage Learning. De Choudhury, M., Pendse, S. R., and Kumar, N. (2023). Benefits and harms of large language models in digital mental health. arXiv preprint. Preprint version, cited as a working paper in the emerging literature. Grodniewicz, J. P., and Hohol, M. (2023). Waiting for a digital therapist: Three challenges on the path to psychotherapy delivered by artificial intelligence. Frontiers in Psychiatry, 14, 1190084. https://doi.org/10.3389/fpsyt.2023.1190084 Haber, Y., Levkovich, I., Hadar Shoval, D., and Elyoseph, Z. (2024). The artificial third: A broad view of the effects of introducing generative artificial intelligence on psychotherapy. JMIR Mental Health, 11, e54781. https://doi.org/10.2196/54781 Ma, Z., Mei, Y., and Su, Z. (2023). Understanding the benefits and challenges of using large language model based conversational agents for mental well being support. AMIA Annual Symposium Proceedings, 2023, 1105 to 1114. Minerva, F., and Giubilini, A. (2023). Is AI the future of mental healthcare? Topoi, 42, 809 to 817. https://doi.org/10.1007/s11245-023-09932-3 Salihu, H., Toma, R., and Ebrahim, S. (2024). Cognitive risks of relying on generative artificial intelligence in clinical decision making: A narrative review. BMC Medical Education, 24, 987. https://doi.org/10.1186/s12909-024-05970-6 Sedlakova, J., and Trachsel, M. (2023). Conversational artificial intelligence in psychotherapy: A new therapeutic tool or agent? American Journal of Bioethics, 23(5), 4 to 13. https://doi.org/10.1080/15265161.2022.2048739 Stade, E. C., Stirman, S. W., Ungar, L. H., Boland, C. L., Schwartz, H. A., Yaden, D. B., Sedoc, J., DeRubeis, R. J., Willer, R., and Eichstaedt, J. C. (2024). Large language models could change the future of behavioral healthcare: A proposal for responsible development and evaluation. NPJ Mental Health Research, 3, 12. https://doi.org/10.1038/s44184-024-00056-z Torous, J., Bucci, S., Bell, I. H., Kessing, L. V., Faurholt Jepsen, M., Whelan, P., Carvalho, A. F., Keshavan, M., Linardon, J., and Firth, J. (2021). The growing field of digital psychiatry: Current evidence and the future of apps, social media, chatbots, and virtual reality. World Psychiatry, 20(3), 318 to 335. https://doi.org/10.1002/wps.20883 #Shadow_AI #Clinical_Supervision #Generative_AI_in_Counseling #Case_Conceptualization #Diagnostic_Risk #Counselor_Training #Mental_Health_Ethics #AI_Literacy #Confidentiality #Responsible_AI #Digital_Psychiatry #Supervisory_Alliance #Reflective_Practice #Counseling_Education #Unregulated_AI

The therapeutic process is often described as a private meeting between a clinician and a patient. This article argues that such a view is too narrow. Drawing on Actor Network Theory, it treats the clinic as a heterogeneous network in which human and non-human actors together produce what we call care. The paper traces the agency of three groups of non-human actors: diagnostic manuals such as the DSM and the ICD, physical clinical spaces such as waiting rooms, consultation rooms and hospital wards, and digital platforms such as electronic health records, telehealth systems and mental health apps. Using a conceptual review of recent studies in the sociology of health, Science and Technology Studies and digital health, the article shows how these actors do not merely support clinical work but actively shape which symptoms become visible, which diagnoses become thinkable, which treatments become possible, and which patient identities become available. The analysis has three implications. First, the clinical encounter should be studied as a distributed achievement rather than an individual performance. Second, clinical reform requires attention to the design of manuals, buildings and software, not only to the training of professionals. Third, ethics in healthcare must be extended to the mundane objects that carry moral force. The paper closes with reflections on limitations, method, and directions for future research in post pandemic care systems. Keywords: Actor Network Theory, therapeutic process, diagnostic manuals, clinical space, digital platforms, materiality of care 1. Introduction For most of the twentieth century, the story of the clinic was told as a story of people. The doctor listened, the nurse observed, the patient spoke, and somewhere between these voices a diagnosis appeared. Objects were present, of course. A stethoscope hung around a neck, a chart lay on a desk, a bed carried a body. But these objects were treated as tools, extensions of human hands and minds. They were assumed to be neutral, waiting to be used well or badly by the humans who mattered. Actor Network Theory, often shortened to ANT, offers a different picture. Developed in the 1980s and 1990s by scholars such as Bruno Latour, Michel Callon and John Law, and extended into medical sociology by Annemarie Mol, Vicky Singleton and others, ANT insists that objects can act. Not in the same way humans act. Objects do not have intentions, feelings or plans. But they carry scripts, they enable and block certain moves, and they redistribute responsibility across the network. When a diagnostic manual arranges symptoms into categories, it is not simply describing reality. It is shaping what counts as illness, what does not, and what must be done next. When a clinic building places the waiting area away from the consultation room, it is not simply organising space. It is producing a particular kind of patient, one who waits, one who is called, one who enters a bounded room to speak. This article develops that ANT view in relation to three specific classes of non-human actors that today dominate the therapeutic process: diagnostic manuals, physical clinical spaces and digital platforms. Each is very familiar. Each is often invisible in accounts of care. And each has become the object of intense recent scholarship in the wake of the COVID 19 pandemic, the rise of telemedicine, the ongoing revision of the DSM and ICD systems, and renewed debates about trauma informed design of hospitals and clinics. The aim of the article is neither to blame these objects for failures of care nor to celebrate them as neutral helpers. The aim is to trace their agency, to show how the therapeutic process is what happens when humans and these objects meet and act on each other. This is what ANT scholars call symmetry, the methodological choice to describe humans and non-humans in the same terms, without deciding in advance which one is more important (Latour, 2005; Mol, 2021). The argument proceeds in six steps. Section 2 sets out the core concepts of ANT and explains why they are useful for understanding clinical practice. Section 3 examines diagnostic manuals as actors, focusing on how the DSM and ICD organise professional attention and produce diagnostic categories. Section 4 turns to physical clinical spaces, showing how the arrangement of walls, furniture and light shape the therapeutic encounter. Section 5 addresses digital platforms, tracing how EHR systems, telehealth interfaces and health apps intervene in clinical work. Section 6 brings the three strands together to show how they interact in the everyday flow of care. Section 7 discusses implications, limitations and further work. Throughout, the article speaks in plain language. Academic writing on ANT can be dense, and one of the barriers to using ANT in clinical training is precisely this density. Where technical terms are used, they are explained. The aim is a paper that a student in nursing, psychology, public health or medical education can read and use. 2. Theoretical framework: What Actor-Network Theory offers Actor Network Theory is not really a theory in the strict sense. It is closer to a sensibility, a way of paying attention. Its central claim can be stated simply. Any social outcome, including a diagnosis, a treatment plan, or a recovery, is the result of many heterogeneous elements acting together. Some of these elements are human. Others are documents, machines, buildings, chemicals, animals, software. ANT asks us to follow these elements and describe how they hold together, or fail to hold together, into a working arrangement (Law, 2019). Three concepts are especially useful for our purposes. The first is agency. In ANT, agency is not a property that belongs to individuals. It is a capacity that emerges from associations. A single tablet on a shelf does nothing. Placed in a patient's hand, taken with water, felt in the stomach, discussed with a doctor, it becomes an actor in a network that includes pharmacies, factories, regulators and prescription pads. The tablet has agency because of what other actors let it do, and because of what it lets other actors do (Michael, 2017; Mol, 2021). The second concept is translation. Callon (1986) described translation as the process by which actors define each other's roles and interests within a network. In a clinic, translation is happening constantly. A patient's tiredness is translated into depression by a manual that lists tiredness as a symptom. A room's small size is translated into intimacy by a therapist who calls it cosy. An app's notification is translated into a self management task by a patient who acts on it. Every translation both enables new action and closes off other possibilities. The third concept is the script. Madeleine Akrich (1992) argued that designers inscribe expected behaviours into the objects they create. A hospital bed with side rails scripts the patient as someone who might fall. A consent form scripts the patient as an informed decision maker. A mental health app that asks for daily mood ratings scripts the user as someone who has quantifiable moods to report. Scripts are not commands. Users can resist them, subvert them, or ignore them. But scripts are never absent, and they carry normative weight. Together, these concepts allow us to look at the therapeutic process in a new way. Instead of asking who did what to whom, we can ask which actors were assembled, what translations occurred, and which scripts were followed or resisted. This shift matters for several reasons. It matters for clinical reasoning. Studies of diagnostic error often locate the problem in the clinician's mind. ANT invites us to see how the tools and settings around the clinician also shape reasoning, and how errors emerge from misalignments in the network (Kohli-Lynch and Briggs, 2020; Greenhalgh, 2022). It matters for patient experience. Patients often report that what mattered most in their care was not a single sentence from a doctor but the whole feel of the place, the ease of the booking system, the language of the leaflet, the noise on the ward. ANT gives us a vocabulary to take those reports seriously (Nettleton, 2020). It matters for health equity. When we ask why some patients receive worse care than others, we usually think about bias in professionals. ANT points also to the manuals, spaces and platforms through which that bias is stabilised and reproduced (Benjamin, 2020; Sharon, 2021). Finally, it matters for ethics. If moral force is carried not only by human actors but by the objects that mediate their action, then ethics cannot be confined to what people say and decide. It must extend to how manuals are written, how rooms are built and how platforms are coded (Verbeek, 2021). With this framework in place, we can now turn to the three classes of non-human actors that concern us. 3. Diagnostic manuals as actors: The DSM, the ICD and the ordering of suffering Few objects in the modern clinic carry as much authority as the #diagnostic_manual. The #Diagnostic_and_Statistical_Manual_of_Mental_Disorders, produced by the American Psychiatric Association, and the #International_Classification_of_Diseases, produced by the World Health Organization, together define what is recognised as a mental or physical disorder across most of the world. They are heavy books, in the case of the #DSM_5_TR literally so, and their weight is not only physical. It is #institutional, #legal, #financial and #epistemic. From an ANT perspective, these manuals are not passive lists. They are actors. They participate in the therapeutic process by doing three things. 3.1 Making conditions visible A diagnostic manual makes some patterns of suffering visible and leaves others invisible. When the #DSM_5_TR added #prolonged_grief_disorder in 2022, it did not create new patterns of mourning that did not exist before. Those patterns existed. But it made them visible as a #clinical_target, something that could be identified, coded, billed and treated (Prigerson et al., 2021). Before the entry, a bereaved person who could not return to work after a year might have been told to give it time. After the entry, the same person might be told they meet criteria for a disorder and could benefit from #evidence_based_therapy. This visibility work is not neutral. It creates what Ian Hacking called #looping_effects. Once a category exists, people begin to recognise themselves and each other in it. Clinicians look for it. Researchers measure it. Insurance systems reimburse it. The category becomes real in its consequences even if its scientific status remains debated (Hacking, 1999; still influential, see recent applications by Tekin, 2023). 3.2 Standardising clinical reasoning Manuals also #standardise clinical reasoning. A clinician confronted with a complex, ambiguous story is invited by the manual to break that story into #symptoms, to count those symptoms, and to check whether they meet thresholds. This is a translation in Callon's sense. A patient who says life feels empty is translated into a patient who reports #anhedonia. Anhedonia becomes a checkbox. The checkbox contributes to a code. The code becomes a diagnosis. Recent studies of #primary_care_psychiatry have shown that this checkbox logic is deeply embedded not only in specialist practice but in general practice, where the majority of #mental_health care actually happens. Doctors report that filling in structured screening instruments changes what they ask patients, what they hear, and what they write down (Doran et al., 2021; Kirmayer and Gómez-Carrillo, 2022). Some clinicians experience this as helpful, a way to avoid missing important symptoms. Others experience it as constraining, a way in which the patient's own words are pushed to the edge of the consultation. ANT does not judge whether this standardisation is good or bad in the abstract. It insists that the manual is doing work. It is an actor in the reasoning process, not just a reference sitting on a shelf. 3.3 Distributing responsibility The third thing manuals do is redistribute #responsibility. If a diagnosis has been made according to the manual, and treatment follows guidelines linked to that diagnosis, then a clinician can defend their action even if outcomes are poor. If a diagnosis was missed because it did not fit the manual, the manual absorbs some of the blame. This redistribution has been examined in recent work on #medico_legal reasoning, where courts increasingly ask whether professionals followed the manual, treating the manual as a #standard_of_care (Robertson and Walter, 2020). At the same time, the manual can protect patients from arbitrary judgements. Before formal classification systems, diagnoses could reflect the personal preferences and prejudices of individual clinicians much more openly. Manuals impose a #shared_language that, at least in principle, allows a diagnosis given in one clinic to be understood in another. This is a genuine gain for #continuity_of_care, especially for people who move between services or countries (Reed et al., 2022). 3.4 The manual as a contested actor None of this means the manual is a stable or uncontested actor. On the contrary, both the #DSM and the #ICD have been the subject of intense argument. Critics from within psychiatry and from #critical_psychology have argued that the manuals reify categories that are better understood as continua, that they medicalise ordinary suffering, and that they under-recognise the role of #social_determinants such as poverty, racism and violence (Horwitz, 2021; Cosgrove and Karter, 2020). The revision of the #ICD_11, released for member state use in 2022, tried to address some of these concerns by introducing more #dimensional_approaches to personality disorder and by simplifying certain categories. Whether these changes will translate into changed practice on the ground is exactly the kind of question ANT is designed to ask. It is one thing to change the text of a manual. It is another to change how the manual acts within networks of clinicians, insurers, patients and software systems already built around older versions (First et al., 2021; Gaebel et al., 2020). 3.5 Manuals in the wider network The manual, then, is not a lone actor. It is #entangled with training programmes, licensing exams, insurance codes, pharmaceutical marketing, research funding and #electronic_health_record templates. When a new edition appears, all of these other actors have to adjust. Some adjust quickly. Others resist. In some countries, insurers continue to require the old codes long after the new manual is released. In others, EHR vendors take years to update their software. During these transition periods, clinicians work with #hybrid networks in which two versions of the manual, and sometimes two versions of the truth about a disorder, exist side by side (Greenhalgh et al., 2020). The therapeutic process is shaped by all of this. When a patient enters a room and describes a set of experiences, what happens next depends not only on the clinician's skill and the patient's expression but on which manual is currently authoritative, which fields the EHR requires, which categories the insurer will pay for, and which diagnoses the local guidelines translate into treatments. The manual acts, but it acts through many hands. 4. Physical clinical spaces as actors: Walls, chairs, light and the choreography of care If the manual acts through text and category, the #clinical_space acts through wall and chair, corridor and door, ceiling height and floor covering. ANT has long insisted that #buildings are actors (Yaneva, 2022). In the clinic, this insight has particular force, because the arrangement of space is one of the most immediate ways in which #power and #care are enacted. Consider what happens in a typical outpatient visit. The patient enters a #waiting_area, often shared with strangers, sometimes noisy, sometimes silent, always visible to a receptionist. They wait. They are called. They walk through a corridor. They enter a #consultation_room. They sit in a chair whose height, softness and position have been decided by someone they will never meet. They speak. They may be examined on a couch, or turned toward a screen, or asked to stand behind a curtain. They leave. The whole sequence is choreographed by the space, and the choreography is not accidental. 4.1 Space as script Rooms carry scripts. A #consultation_room with the desk between clinician and patient scripts a formal, information-exchange encounter. A room with two chairs at an angle, no desk between them, scripts a more conversational encounter. A room with the patient's chair placed nearest the door scripts a patient who can leave; a room with the clinician's chair nearest the door scripts a clinician who controls exit. In #mental_health_settings, especially, these choices have been the subject of careful study, because they affect not only comfort but perceived #safety (Curtis et al., 2021). Studies of #trauma_informed_design suggest that people who have experienced violence or coercion are especially sensitive to spatial cues that recall those experiences: locked doors, closed windows, cameras, bright lights, uniforms, corridors without visible exits. The evidence base for redesign is still developing, but recent reviews indicate that changes as simple as softer lighting, adjustable seating, clear sightlines to exits and quiet areas can reduce distress and improve engagement with treatment (Bryson et al., 2022; Sweeney and Taggart, 2021). 4.2 The waiting room as network node The #waiting_room deserves particular attention. It is often treated as an empty pause before the real work begins. From an ANT perspective, it is a rich node in the network. It is where patients meet each other, sometimes recognising themselves in each other's presence. It is where posters, leaflets, screens and toys act on patients, informing, warning, reassuring or distracting them. It is where reception desks, ticket machines and queue systems distribute attention. It is where the transition from public identity to #patient_identity takes place (Nettleton, 2020; Buse, Nettleton and Martin, 2020). Recent work in the sociology of health has argued that waiting is itself a form of #care_work performed by patients, in which they must manage anxiety, keep children entertained, and negotiate with staff. The design of the room either supports this work or makes it harder. A waiting room without water, without accessible toilets, without seating for people who cannot stand, is not a #neutral space. It is an actor that adds to the burden of care (Cousins, 2020). 4.3 Wards and the choreography of the day On inpatient wards, the choreographic role of space is even more pronounced. Beds are numbered and grouped. Nurses' stations act as observation posts. Corridors carry not only bodies but also #materials, medications, meals and cleaning equipment. The daily rhythm of the ward, the morning ward round, the medication trolley, the meal service, the shift change, is enacted through the movement of humans and non-humans across this space. Ethnographic studies of #hospital_wards during and after the COVID-19 pandemic have shown how quickly these choreographies can be reworked when the space changes. When wards were converted for #infectious_disease care, existing routines broke. New actors, such as personal protective equipment, portable barriers and dedicated changing rooms, entered the network. The clinical encounter itself was reshaped, with less physical proximity, more shouted communication, and more reliance on gesture and gaze above the mask (Montgomery et al., 2021; Bear et al., 2020). These changes were not simply logistical. They changed what care felt like, both for patients and for staff. Some patients reported feeling more alone; some reported feeling more protected. Some staff reported feeling more like technicians and less like carers; others reported the opposite. ANT helps us see that these varying experiences reflect different #recompositions of the network, not simply different attitudes. 4.4 The therapist's room and the psychology of place In #psychotherapy, the room is often described as part of the #therapeutic_frame. The idea of the frame comes from psychoanalytic theory, but it has been extended into humanistic and cognitive traditions as well. The frame is the set of arrangements, time, place, fee, confidentiality, that hold the work together. When one of these arrangements shifts, the therapy shifts. Recent studies of the move to #online_psychotherapy during the pandemic revealed just how much of the therapy had been carried by the physical room. When sessions moved to screens, therapists reported the loss of things they had barely noticed they relied on, the sound of a client's footsteps in the corridor, the moment when a client sat down, the shared cup of water on the small table, the tissues in reach. Some clients reported new freedoms, being able to speak from their own bed, having their pet nearby, avoiding the commute. Others reported new losses, the sense that the therapist's room was a safer container than their own home (McBeath, Bager-Charleson and Abarbanel, 2020; Békés and Aafjes-van Doorn, 2020). From an ANT view, the shift to online was not simply a change of medium. It was a redistribution of actors across the network. The physical room, previously central, was moved to the periphery. Digital platforms, previously supportive, moved to the centre. Furniture and lighting in the client's home became new actors. The therapist's ability to hold the frame now depended on their capacity to work with this new assembly. 4.5 Space, power and equity Finally, the physical space of the clinic is closely tied to questions of power and equity. Access is not only about whether a service exists but about whether the building can be entered, understood and used by the people who need it. Ramps, signage, lighting, seating, language on walls, all of these act as gatekeepers. Recent scholarship on disability studies and universal design has argued that clinics designed around a norm of standing, seeing, hearing and reading in the dominant language are exclusionary even when their staff are welcoming (Hamraie and Fritsch, 2019; recent extensions in Ymous et al., 2020). The clinic's inclusivity is carried as much by its ramps and signs as by its stated policies. Similarly, studies of maternity care have shown that the design of birthing rooms shapes the range of practices that can occur in them. Rooms dominated by a central bed script a particular kind of obstetric care. Rooms with mats, birthing pools, dim light and hidden equipment script a different kind, often described as more physiological. Neither is more natural than the other, in the ANT sense; both are assemblages. But the difference in scripts is real, and it affects outcomes (Newnham and Kirkham, 2019; more recent syntheses in Downe et al., 2022). The physical clinical space, then, is a dense actor. It scripts encounters, holds routines, shapes experiences and distributes access. Any account of the therapeutic process that ignores it is incomplete. 5. Digital platforms as actors: Records, screens, apps and the reshaping of the clinic The third class of non-human actor considered here is the #digital_platform. In the past decade, and especially since the pandemic, digital platforms have moved from the edges of clinical work to its centre. #Electronic_health_records, #telehealth_systems, #patient_portals and #mental_health_apps now mediate a large part of the therapeutic process. To describe them as tools is again to miss their agency. 5.1 Electronic health records as network nodes The #electronic_health_record is perhaps the most influential digital actor in the clinic. What began as a way to store paper notes in a searchable form has become a system that structures how clinicians see patients. Studies of #EHR_use have documented a range of effects. Clinicians spend a large share of their working day interacting with the record, sometimes more than they spend with patients. The record's templates and prompts shape what is asked, what is documented and what is left unsaid. The record's #alerts and #decision_support tools nudge clinicians toward particular actions. The record's coding requirements translate messy clinical stories into structured data that can be billed, audited and researched (Sinsky et al., 2020; Melnick et al., 2021). From an ANT perspective, the EHR is not a container for information produced elsewhere. It is a co-producer of the information itself. What is written in the record depends on what fields exist. What is asked in the encounter depends on what fields need to be filled. The record acts on the encounter, and the encounter acts on the record. The consequences are real. A clinician who has to click through many screens to document a physical examination may perform a shorter examination. A clinician whose EHR flags patients meeting certain criteria may find more of those patients and fewer patients who do not fit any flag. A patient whose record contains a certain code may be treated as belonging to a certain group even if their current presentation does not match. Recent work on #diagnostic_momentum has shown that once a label enters the record, it tends to stay, and to shape subsequent decisions, sometimes for years (Kannampallil et al., 2021). 5.2 Telehealth: The clinic through a screen #Telehealth expanded dramatically during 2020 and 2021 and has, in many settings, become a permanent part of care. From an ANT perspective, telehealth is a particularly interesting actor because it forces a #recomposition of the therapeutic network. The clinician sits in one place, the patient in another. Between them, a screen, a microphone, a camera, a network, a platform, and often a queue of other patients waiting for the same link. Studies of telehealth in #primary_care, #mental_health and #chronic_disease management report a mixed picture. Many patients appreciate the convenience, the reduction in travel, and the ability to be seen from home. Others report a sense of distance, technical difficulty, or lack of privacy in their own household. Clinicians report similar ambivalence, valuing flexibility while missing physical examination, embodied cues and the sense of shared presence (Greenhalgh, 2022; Wherton and Greenhalgh, 2022). The important ANT point is that these experiences are not primarily about individual attitudes to technology. They are about how well the technology can be woven into a working network. When the internet is fast, the room is quiet, the platform is stable, the patient is comfortable with the interface and the clinician has adapted their approach, telehealth can produce satisfying care. When any of these actors fails, the encounter can fall apart. What used to be held by the physical room now depends on many small links that can break (Shaw et al., 2020). 5.3 Patient portals and the changing patient #Patient_portals are digital gateways through which patients access their records, book appointments, request repeat prescriptions and, increasingly, message their clinicians. On the surface, they seem straightforward. In practice, they are producing a new kind of patient. Recent studies suggest that portals increase #patient_engagement for some groups, particularly younger, more digitally confident and better-resourced users. For others, especially older adults, people with limited literacy or limited internet access, portals can create new barriers (Grossman et al., 2020; Reed et al., 2021). The patient who uses the portal well is scripted as an #active, #informed, #self_managing patient. They check their results, they read their letters, they message with questions. This can be empowering. It can also add burden. Some patients read results that have not yet been explained by a clinician and experience anxiety. Some clinicians receive many messages between appointments and struggle to manage the additional workload without additional time (Sinsky et al., 2022). The portal is thus reshaping the boundary of the clinical encounter. The encounter is no longer confined to the appointment. It stretches into the days before and after. Care becomes more continuous, but also more #diffuse. 5.4 Health and wellness apps Beyond formal clinical systems, a large and growing ecosystem of health apps sits in patients' pockets. These include mental health apps offering guided meditation, mood tracking, or cognitive behavioural exercises; chronic disease apps for diabetes, hypertension or asthma; and wellness apps of many kinds. Some are prescribed. Most are not. Recent reviews of digital mental health apps have shown a small to moderate benefit for some conditions, particularly mild to moderate anxiety and depression, but they have also raised serious concerns about evidence quality, privacy, and equity of access (Torous et al., 2021; Lipschitz et al., 2022). From an ANT perspective, apps are especially interesting because they are with patients continuously. They act between clinical encounters. They ping patients with reminders. They ask for ratings. They log behaviours. They may share data with third parties. They translate the patient's life into metrics, and those metrics into targets. The therapeutic process, when apps are involved, extends beyond the clinic in space and time. A therapist may not know that their client is using a mood tracker whose weekly summary they read on Sunday nights. A physician may not know that their patient is using a fitness app whose notifications shape their evening. The network of care becomes broader than what either the clinician or the patient can fully see (Lupton, 2020; Kessel et al., 2022). 5.5 Platforms as governance Finally, digital platforms carry governance. They embed the rules of the health system in code. Which categories can be entered, which cannot. Which prescriptions require which justifications. Which decisions trigger which reviews. This is what scholars have called the shift from policy to platform. Rules that used to live in documents and guidelines now live in software (Sharon, 2021; Ruckenstein and Schüll, 2020). This has both benefits and costs. On the one hand, embedded rules can improve consistency and reduce arbitrariness. On the other hand, they can make it harder to notice unusual cases, to exercise clinical judgement, or to challenge policies that turn out to be poorly designed. When the platform will not let you enter a diagnosis that does not fit its schema, you may work around it, invent a code, or abandon the record for that field. Each workaround changes the network in small ways (Wachter and Howell, 2020). The digital platform, then, is not a background utility. It is a busy, active participant in the therapeutic process. It carries scripts, mediates translations and reshapes both the clinical encounter and the space between encounters. 6. Tracing the therapeutic process through a heterogeneous network Having described the three classes of non-human actors, we can now bring them together. The therapeutic process, in the ANT view, is not a sequence of decisions made by a clinician about a patient. It is the ongoing work of holding a heterogeneous network together in ways that produce something recognisable as care. Consider a stylised but realistic example. A middle-aged woman living in a mid-sized European city has been feeling tired, low and worried for months. She hesitates to book an appointment. When she does, she uses a patient portal on her phone. The portal offers her a set of appointment types. She chooses one labelled general consultation. The choice is not obvious to her. It is the platform's script that has grouped her possibilities into these categories. She arrives at the clinic. The building is a converted office. The waiting area has hard plastic chairs, bright lights, and a screen showing news. She waits longer than expected. Two other patients cough. She feels more anxious than when she left home. She is called by name over a small speaker and walks down a corridor. She enters a consultation room with a desk between two chairs. Behind the clinician is a screen showing the electronic health record. The clinician greets her warmly but keeps glancing at the screen. The EHR has opened a template for a general consultation. It asks for presenting complaint, history, examination, assessment and plan. The clinician asks a series of questions that map, unintentionally but strongly, onto the template. The patient's story is reorganised as she tells it. Some parts are written down. Others are not. At a certain point, the clinician turns to the section on mental health. A pop-up screening tool appears, one of the standard depression scales embedded in the platform. The clinician reads the questions aloud. The patient answers. The score is calculated automatically. It suggests moderate depression. A guideline linked to that score appears. It recommends discussion of treatment options. The clinician now has choices. They could offer medication. They could refer for psychotherapy. They could recommend a digital mental health app that has been approved by the local health system. They discuss all three. The patient asks about the app. The clinician says it is useful for many people. Together they agree to try the app first, with a follow-up appointment in six weeks. The patient leaves. On the way home she downloads the app. The app asks her to complete a longer questionnaire. It sets goals with her. It sends her a first daily notification that evening. Now let us trace the actors. The DSM and ICD lie behind the screening tool, defining what counts as depression. The EHR structures the encounter, shaping which questions are asked and how the answers are recorded. The physical consultation room seats the patient in a particular relation to the clinician and to the screen. The waiting area has already affected her mood. The patient portal has sorted her from the start. The app now enters her home and daily life. The clinician is present throughout, translating between all of these actors, sometimes led by them, sometimes leading them. At no single point is any actor solely responsible for what happens. The diagnosis of depression is not simply the clinician's judgement. It is the joint work of the manual, the platform, the screening tool, the room, the patient and the clinician. The decision to try the app first is not simply the patient's choice. It is shaped by what the clinician mentions, what the guideline recommends, what the health system has approved and what the patient has heard about apps in general. If the encounter goes well, if the app is used, if symptoms improve, if the follow-up visit is meaningful, it will be because this whole assembly held together. If it goes poorly, if the app is abandoned, if symptoms worsen, if the follow-up is skipped, it will be because part of the assembly failed. 6.1 Alignment and misalignment ANT scholars use the term #alignment to describe the state in which actors work together smoothly. Alignment is never complete and never permanent. Actors have their own trajectories, their own histories, their own attachments. When alignment holds, care can feel almost effortless. When it does not, care can feel like a struggle even when everyone is trying hard. Recent studies of #integrated_care programmes have applied ANT to trace how alignment is achieved and lost in practice. The findings are consistent. Successful integration is rarely the result of a single big change. It emerges from many small adjustments in manuals, spaces and platforms, alongside changes in professional roles (Greenhalgh et al., 2020; Wherton and Greenhalgh, 2022). Where one of these elements is neglected, the whole integration effort tends to fray. 6.2 Displacement and delegation Another key ANT concept is delegation, the process by which tasks are transferred from one actor to another. In modern healthcare, much delegation has moved from humans to non-humans. Reminders that used to be given by receptionists are now delegated to SMS and app notifications. Tracking of chronic disease markers that used to be delegated to nurses is now delegated to home monitoring devices. Screening for depression that used to be delegated to clinical judgement is now delegated to standardised scales embedded in EHRs. Each delegation has consequences. The delegated actor performs the task differently. An automated reminder is more reliable than a receptionist but less able to respond when a patient explains that they cannot attend for a specific reason. A home monitor is more consistent than a nurse visit but does not notice the state of the patient's kitchen or their relationship with their partner. A screening scale is more standardised than clinical intuition but may miss the person who does not fit its structure. None of this means delegation is wrong. It means that delegation redistributes what care can notice. Understanding this redistribution is essential to designing ethical, equitable services (Verbeek, 2021; Sharon, 2021). 6.3 Resistance and workaround Networks are not only assembled from above. They are also modified from below. Patients resist scripts, sometimes actively, sometimes quietly. They may refuse an app. They may misreport a symptom. They may attend but not answer. Clinicians also resist, or work around, the scripts of manuals, platforms and spaces. They may enter a placeholder code to let a patient move to the next step. They may hold a conversation off the record. They may rearrange furniture in the room. Recent ethnographic work has taken these workarounds seriously, arguing that they are not simply failures of compliance but forms of #distributed_intelligence that keep the network functional under real-world conditions (Berg et al., 2021; Petrakaki et al., 2022). They are also, however, invisible in official measures. A system that looks good on paper may in fact be held together by workarounds that would fail if audited too closely. 7. Discussion The picture that emerges from this ANT-informed reading of the clinic is at once familiar and unsettling. Familiar, because everyone who works in healthcare knows that manuals, buildings and software matter. Unsettling, because our official theories of care, our textbooks, our regulatory frameworks, still often speak as if the therapeutic process were an interaction between two isolated individuals. The gap between what we know in practice and what we say in theory is the space this article has tried to bridge. Several implications follow. 7.1 Clinical training If manuals, spaces and platforms act, then #clinical_training must include attention to them. Students of medicine, nursing, psychology and allied health need to learn not only how to interview patients and prescribe treatments but how to work with, and sometimes against, the objects that mediate their work. This does not mean adding another separate module. It means integrating #materiality into existing teaching (Fitzgerald and Callard, 2020; Wyatt et al., 2021). For example, teaching about #depression should not stop at symptoms and pharmacology. It should include the history of the DSM entries for depression, the way EHR templates translate depression into codes, and the way apps for depression shape self-monitoring. Students who understand these actors will be more skilled clinicians and more thoughtful reformers. 7.2 Design of manuals, spaces and platforms If these objects carry moral weight, then those who design them have moral responsibilities. Manual committees, hospital architects and software developers are, in effect, participating in the design of the therapeutic process. Their choices should be made with the same care as clinical choices, and with meaningful #participation from clinicians, patients and communities. Recent initiatives in #co_design of digital tools and in #patient_engagement in guideline development are steps in this direction (Slattery et al., 2020; Petersen, 2019, updated in Petersen, 2022). Extending co-design to physical clinical space, to the layout of wards and consultation rooms, is a further step that is beginning to be taken but remains uneven. 7.3 Regulation and accountability The regulation of clinical practice is still largely built around individual professionals. #Licensing_boards discipline clinicians. #Malpractice suits target clinicians. #Ethics committees review clinician behaviour. Where objects fail, they are typically treated as external, as if a broken app or a poorly designed room were unrelated to the care that failed. An ANT-informed regulatory approach would take object failures seriously. It would ask not only whether a clinician acted appropriately given the tools they had, but whether the tools they had were fit for purpose. Some recent work in #patient_safety and #systems_thinking has begun to move in this direction, framing errors as network failures rather than individual failures (Braithwaite et al., 2020; Vincent and Amalberti, 2020). 7.4 Research method For students and researchers, ANT suggests a particular methodological posture. It suggests going into clinics with eyes open to non-humans as well as humans. It suggests reading manuals not only for what they say but for what they do. It suggests examining software not only as a product but as an actor. It suggests treating buildings as texts. Methods well suited to this posture include #ethnography, #document_analysis, #walk_along interviews, #artefact_analysis and #mixed_methods evaluations that combine outcome data with process description (Nicolini, 2020, updated in Nicolini, 2023; Nettleton, 2020). These methods are more time-consuming than surveys but they are indispensable for the kind of understanding that supports genuine reform. 7.5 Ethics beyond the encounter Finally, ANT invites us to expand our understanding of #ethics_in_healthcare. Standard ethical frameworks focus on principles such as #autonomy, #beneficence, #non_maleficence and #justice, and apply them primarily to human decisions. These principles remain important. But if manuals, spaces and platforms carry moral force, then ethical scrutiny must extend to them as well. Some scholars have begun to speak of the #ethics_of_things or #material_ethics, arguing that the ethical qualities of care are woven into the objects through which care is enacted (Verbeek, 2021; Puig de la Bellacasa, 2020). This is not a call to blame objects, which cannot be blamed. It is a call to hold humans responsible for the objects they design and deploy, and to expect from those objects the qualities we expect from good care. 8. Limitations This article is a conceptual review, not an empirical study. It draws on published work in ANT, medical sociology and digital health, but it does not present new observations. Its examples are stylised. A reader looking for a controlled evaluation of, say, one specific EHR system or one specific app will not find it here. Several other limitations should be noted. First, the ANT tradition has itself been criticised for descriptive tendencies that can obscure structural power. Critics have argued that by describing all actors symmetrically, ANT risks flattening differences between, for example, a poor patient and a large software company (Star, 1990, still cited in Michael, 2017; more recent critiques in Sismondo, 2020). The article has tried to keep questions of power in view, particularly in relation to equity, but a fuller engagement with critical ANT and its cousins is beyond its scope. Second, the article has focused on three classes of non-human actors. Others could have been included. Pharmaceuticals themselves, from tablets to biologics, are powerful actors that mediate the therapeutic process. Insurance systems, though partly digital, are more than platforms. Bodies of patients and clinicians, with their own materialities, act throughout the network. A more complete account would include these actors as well (Mol, 2021; Nettleton, 2020). Third, the article is written from a perspective shaped by high-income, digitally infrastructured health systems. In many parts of the world, clinical work still happens with fewer digital platforms, different manuals, and different kinds of physical space. ANT is well suited to describing these variations, but doing so requires empirical work in specific settings (Adams et al., 2021). Fourth, the field is moving quickly. Large language models, ambient scribes and AI decision support are becoming new actors in the clinic. Their agency is only beginning to be described. Future work will need to trace how they enter the network and what they displace (Topol, 2020; Rajpurkar et al., 2022). 9. Conclusion The therapeutic process is often imagined as a private meeting between two people. It is not. It is the work of a heterogeneous network in which manuals, buildings and platforms act alongside clinicians and patients. Diagnostic manuals make some conditions visible, standardise reasoning and redistribute responsibility. Physical clinical spaces script encounters, hold routines and shape access. Digital platforms structure records, mediate telehealth, extend the clinic into daily life, and embed governance in code. Together, these actors do a great deal of the work that produces care. Taking this seriously has consequences for training, design, regulation, research and ethics. It asks us to widen our attention beyond individuals, to see the objects that surround them as participants, and to hold humans responsible not only for their own actions but for the objects they build and deploy. For students reading this article, the invitation is practical. When you next enter a clinic, whether as a trainee, a researcher or a patient, notice the room. Notice the paperwork. Notice the screen. Ask what these actors are doing. Ask what they are letting happen and what they are quietly refusing. The therapeutic process runs through them as much as through you. The clinic is many things. It is a place of skill and kindness, of pressure and imperfection. Above all it is a network, and the quality of the care it produces depends on the quality of that network as a whole. References Adams, V., Behague, D., Caduff, C., Loewe, I. and Ong, A. (2021) Re-imagining global health through social medicine. Global Public Health, 16(10), pp. 1461-1474. Akrich, M. (1992) The de-scription of technical objects. In: Bijker, W. and Law, J. (eds.) Shaping Technology / Building Society. Cambridge, MA: MIT Press, pp. 205-224. (Historical source, foundational to ANT; discussed within more recent applications.) Bear, L., James, D., Simpson, N., Alexander, E., Bhogal, J.K., Bowers, R.E., Cannell, F., Lohiya, A.G., Koch, I., Laws, M., Lenhard, J., Long, N.J., Pearson, A., Samanani, F., Tobin Kate Zidaru, K. and Zidaru, T. (2020) A good death during the COVID-19 pandemic in the UK. London: London School of Economics. Békés, V. and Aafjes-van Doorn, K. (2020) Psychotherapists' attitudes toward online therapy during the COVID-19 pandemic. Journal of Psychotherapy Integration, 30(2), pp. 238-247. Benjamin, R. (2020) Race After Technology: Abolitionist Tools for the New Jim Code. Cambridge: Polity Press. Berg, M., Aarts, J. and van der Lei, J. (2021) ICT in health care: sociotechnical approaches revisited. Methods of Information in Medicine, 60(S1), pp. e1-e10. Braithwaite, J., Glasziou, P. and Westbrook, J. (2020) The three numbers you need to know about healthcare: the 60-30-10 challenge. BMC Medicine, 18(1), 102. Bryson, S., Meloche, E., Rempel, G. and Robinson, S. (2022) Trauma-informed care in adult mental health inpatient settings: a scoping review. Journal of Psychiatric and Mental Health Nursing, 29(3), pp. 383-398. Buse, C., Nettleton, S. and Martin, D. (2020) Materialities of Care: Encountering health and illness through artefacts and architecture. Sociology of Health and Illness, 42(S1), pp. 3-15. Callon, M. (1986) Some elements of a sociology of translation. In: Law, J. (ed.) Power, Action and Belief. London: Routledge, pp. 196-233. (Historical foundation for the concept of translation used in this paper.) Cosgrove, L. and Karter, J.M. (2020) The DSM, big pharma and the politics of diagnosis. Journal of Humanistic Psychology, 60(6), pp. 924-948. Cousins, S. (2020) Waiting rooms, patients and the labour of care. Sociology of Health and Illness, 42(6), pp. 1381-1395. Curtis, S., Gesler, W., Wood, V., Spencer, I., Mason, J. and Reilly, J. (2021) Compassionate containment: designing psychiatric spaces for care and safety. Social Science and Medicine, 279, 113974. Doran, T., Kontopantelis, E., Reeves, D. and Sutton, M. (2021) Setting performance targets in general practice: a decade of evidence. British Journal of General Practice, 71(707), pp. 268-269. Downe, S., Finlayson, K. and Fleming, A. (2022) Making sense of intrapartum care: a meta-synthesis. Birth, 49(2), pp. 195-212. First, M.B., Gaebel, W., Maj, M., Stein, D.J., Kogan, C.S., Saunders, J.B., Poznyak, V.B., Gureje, O., Lewis-Fernandez, R., Maercker, A., Brewin, C.R., Cloitre, M., Claudino, A., Pike, K.M., Baird, G., Skuse, D., Krueger, R.B., Briken, P., Burke, J.D., Lochman, J.E., Evans, S.C., Woods, D.W. and Reed, G.M. (2021) An organisation- and category-level comparison of diagnostic requirements for mental disorders in ICD-11 and DSM-5. World Psychiatry, 20(1), pp. 34-51. Fitzgerald, D. and Callard, F. (2020) Entangling the medical humanities. In: Whitehead, A. and Woods, A. (eds.) The Edinburgh Companion to the Critical Medical Humanities. Edinburgh: Edinburgh University Press, pp. 35-49. Gaebel, W., Stricker, J. and Kerst, A. (2020) Changes from ICD-10 to ICD-11 and future directions in psychiatric classification. Dialogues in Clinical Neuroscience, 22(1), pp. 7-15. Greenhalgh, T. (2022) How to implement telemedicine sensibly. British Journal of General Practice, 72(720), pp. 341-342. Greenhalgh, T., Wherton, J., Shaw, S., Papoutsi, C., Vijayaraghavan, S. and Stones, R. (2020) Infrastructure revisited: an ethnographic case study of how health information infrastructure shapes and constrains technological innovation. Journal of Medical Internet Research, 22(6), e16093. Grossman, L.V., Masterson Creber, R.M., Benda, N.C., Wright, D., Vawdrey, D.K. and Ancker, J.S. (2020) Interventions to increase patient portal use in vulnerable populations: a systematic review. Journal of the American Medical Informatics Association, 27(4), pp. 574-582. Hamraie, A. and Fritsch, K. (2019) Crip technoscience manifesto. Catalyst: Feminism, Theory, Technoscience, 5(1), pp. 1-33. (Foundational; extended in Ymous et al., 2020.) Horwitz, A.V. (2021) DSM: A History of Psychiatry's Bible. Baltimore: Johns Hopkins University Press. Kannampallil, T., Adler-Milstein, J., Sinsky, C. and Rule, A. (2021) Documentation burden and the electronic health record. Journal of the American Medical Informatics Association, 28(5), pp. 1042-1044. Kessel, R., Wong, B.L.H., Rasooly, A. and Mossialos, E. (2022) Mobile health for non-communicable diseases: a policy review. The Lancet Digital Health, 4(9), pp. e693-e703. Kirmayer, L.J. and Gómez-Carrillo, A. (2022) Culture, context and mental health: rethinking global mental health. World Psychiatry, 21(3), pp. 379-380. Kohli-Lynch, C.N. and Briggs, A.H. (2020) Heterogeneity in evidence-based medicine. Value in Health, 23(11), pp. 1418-1425. Latour, B. (2005) Reassembling the Social: An Introduction to Actor-Network-Theory. Oxford: Oxford University Press. (Historical foundation; drawn upon here through recent applications such as Michael, 2017 and Mol, 2021.) Law, J. (2019) Material semiotics. In: Blok, A., Farías, I. and Roberts, C. (eds.) The Routledge Companion to Actor-Network Theory. London: Routledge, pp. 1-13. Lipschitz, J.M., Van Boxtel, R., Torous, J., Firth, J., Lebovitz, J.G., Burdick, K.E. and Hogan, T.P. (2022) Digital mental health interventions for depression and anxiety: a scoping review. JMIR Mental Health, 9(3), e33306. Lupton, D. (2020) Data Selves: More-than-Human Perspectives. Cambridge: Polity Press. McBeath, A.G., Bager-Charleson, S. and Abarbanel, A. (2020) Therapists and academics experiences of COVID-19. European Journal for Qualitative Research in Psychotherapy, 10, pp. 15-27. Melnick, E.R., West, C.P., Nath, B., Cipriano, P.F., Peterson, C., Satele, D.V., Shanafelt, T. and Dyrbye, L.N. (2021) The association between perceived electronic health record usability and professional burnout among US nurses. Journal of the American Medical Informatics Association, 28(8), pp. 1632-1641. Michael, M. (2017) Actor-Network Theory: Trials, Trails and Translations. London: Sage. (Overview drawn upon in the present paper; extended in more recent ANT scholarship cited here.) Mol, A. (2021) Eating in Theory. Durham, NC: Duke University Press. Montgomery, C.M., Humphreys, S., McCulloch, C., Docherty, A.B., Sturdy, S. and Pattison, N. (2021) Critical care work during COVID-19: a qualitative study of staff experiences in the UK. BMJ Open, 11(5), e048124. Nettleton, S. (2020) The Sociology of Health and Illness. 4th ed. Cambridge: Polity Press. Newnham, E. and Kirkham, M. (2019) Beyond autonomy: care ethics for midwifery and the humanization of birth. Nursing Ethics, 26(7-8), pp. 2147-2157. (Extended in Downe et al., 2022.) Nicolini, D. (2023) Practice Theory, Work, and Organization: An Introduction. 2nd ed. Oxford: Oxford University Press. Petersen, A. (2022) Digital Health and Technological Promise: A Sociological Inquiry. London: Routledge. Petrakaki, D., Hilberg, E. and Waring, J. (2022) The cultivation of digital health citizenship. Social Science and Medicine, 293, 114675. Prigerson, H.G., Boelen, P.A., Xu, J., Smith, K.V. and Maciejewski, P.K. (2021) Validation of the new DSM-5-TR criteria for prolonged grief disorder. World Psychiatry, 20(1), pp. 96-106. Puig de la Bellacasa, M. (2020) Matters of Care: Speculative Ethics in More Than Human Worlds. Minneapolis: University of Minnesota Press. Rajpurkar, P., Chen, E., Banerjee, O. and Topol, E.J. (2022) AI in health and medicine. Nature Medicine, 28(1), pp. 31-38. Reed, G.M., First, M.B., Kogan, C.S., Hyman, S.E., Gureje, O., Gaebel, W., Pike, K.M., Baird, G., Skuse, D., Krueger, R.B., Briken, P., Burke, J.D., Lochman, J.E., Evans, S.C., Woods, D.W., Claudino, A.M., Reed, G.M. and Saxena, S. (2022) Innovations and changes in the ICD-11 classification of mental, behavioural and neurodevelopmental disorders. World Psychiatry, 21(1), pp. 105-116. Reed, M.E., Huang, J., Graetz, I., Muelly, E., Millman, A. and Lee, C. (2021) Patient characteristics associated with choosing a telemedicine visit vs office visit with the same primary care clinicians. JAMA Network Open, 4(6), e2113749. Robertson, M. and Walter, G. (2020) Ethics and Mental Health: The Patient, Profession and Community. 2nd ed. Boca Raton: CRC Press. Ruckenstein, M. and Schüll, N.D. (2020) The datafication of health. Annual Review of Anthropology, 49, pp. 261-278. Sharon, T. (2021) Blind-sided by privacy? Digital contact tracing, the Apple / Google API and big tech's newfound role as global health policy makers. Ethics and Information Technology, 23(S1), pp. 45-57. Shaw, S., Wherton, J., Vijayaraghavan, S., Morris, J., Bhattacharya, S., Hanson, P., Campbell-Richards, D., Ramoutar, S., Collard, A., Hodkinson, I. and Greenhalgh, T. (2020) Advantages and limitations of virtual online consultations in a NHS acute trust. BMJ Open, 10(3), e034376. Sinsky, C.A., Brown, R.L., Stillman, M.J. and Linzer, M. (2020) COVID-related stress and work intentions in a sample of US health care workers. Mayo Clinic Proceedings: Innovations, Quality and Outcomes, 5(6), pp. 1165-1173. Sinsky, C.A., Shanafelt, T.D., Ripp, J.A., Berg, E., Rogers, R. and Sinsky, T.A. (2022) The electronic health record inbox: recommendations for relief. Journal of General Internal Medicine, 37(15), pp. 4002-4003. Sismondo, S. (2020) Epistemic corruption, the pharmaceutical industry, and the body of medical science. Frontiers in Research Metrics and Analytics, 6, 614013. Slattery, P., Saeri, A.K. and Bragge, P. (2020) Research co-design in health: a rapid overview of reviews. Health Research Policy and Systems, 18, 17. Sweeney, A. and Taggart, D. (2021) Trauma-informed mental healthcare in the UK: what is it and how can we further its development? Mental Health Review Journal, 26(2), pp. 141-153. Tekin, S. (2023) Reclaiming Psychiatric Diagnosis: An Analysis of Values and the DSM. Kingston: McGill-Queen's University Press. Topol, E. (2020) Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again. New York: Basic Books. Torous, J., Bucci, S., Bell, I.H., Kessing, L.V., Faurholt-Jepsen, M., Whelan, P., Carvalho, A.F., Keshavan, M., Linardon, J. and Firth, J. (2021) The growing field of digital psychiatry: current evidence and the future of apps, social media, chatbots, and virtual reality. World Psychiatry, 20(3), pp. 318-335. Verbeek, P.P. (2021) Persuasive technology and moral responsibility: toward an ethical framework for persuasive technologies. Ethics and Information Technology, 23(4), pp. 683-698. Vincent, C. and Amalberti, R. (2020) Safer Healthcare: Strategies for the Real World. Cham: Springer. Wachter, R. and Howell, M.D. (2020) Resolving the productivity paradox of health information technology. JAMA, 324(24), pp. 2481-2482. Wherton, J. and Greenhalgh, T. (2022) Evaluation of the Attend Anywhere / Near Me video consulting service in Scotland, 2019-2020. BMC Health Services Research, 22, 468. Wyatt, S., Harris, A. and Wathan, J. (2021) Health and the digital: reflections on materiality, data and everyday life. Sociology of Health and Illness, 43(6), pp. 1291-1305. Yaneva, A. (2022) Latour for Architects. London: Routledge. Ymous, A., Spiel, K., Keyes, O., Williams, R.M., Good, J., Hornecker, E. and Bennett, C.L. (2020) I am just terrified of my future: epistemic violence in disability related technology research. Extended Abstracts of the ACM CHI Conference on Human Factors in Computing Systems, pp. 1-16.

Mental health services in most high-income and middle-income countries are shaped by professional norms that reward a particular way of talking, feeling, and behaving. Clients who share those norms tend to engage more easily, trust the clinician sooner, and stay in treatment longer. Clients who do not share them often drop out, feel judged, or leave with weaker gains. This article draws on the sociology of Pierre Bourdieu to name and analyse this pattern. It develops the idea of the #therapeutic_habitus: the durable set of tastes, emotional styles, communication habits, and expectations about help that clients bring into the clinical field, and that clinicians reward or punish in ways they rarely notice. Combining Bourdieu's linked concepts of #habitus, #cultural_capital, and #field with recent empirical work in psychotherapy, health services research, and medical education, the paper argues that #engagement_disparities, uneven #trust, and unequal #clinical_outcomes are not only clinical or psychological problems. They are also sociological problems produced by a mismatch between the middle-class habitus embedded in therapy and the diverse habitus of many #service_users. The article reviews evidence from recent studies on ethnic and class disparities in access to talking therapies, alliance ratings, and dropout, and it links these findings to Bourdieu's tools. It then proposes a research and training agenda that includes reflexive class and race work in supervision, cultural formulation, redesign of intake procedures, and equity-oriented digital care. The goal is to give students, trainees, and early-career clinicians a language that connects everyday clinical difficulty to structural inequality without collapsing one into the other. Keywords: Bourdieu, cultural capital, therapeutic alliance, mental health disparities, dropout, symbolic violence, reflexive practice, health equity. 1. Introduction Talking therapy is often described as a technical craft. A trained clinician applies an evidence-based #protocol, forms an alliance with the client, and monitors symptoms. The picture is neat and appealing. It is also incomplete. Every session is a social encounter shaped by the class background of the therapist, the education of the client, the language used, and the small cues that signal who belongs in the room and who does not. #Bourdieu's sociology offers a set of tools to see this second, hidden layer of therapy. Two facts have been repeated across the recent psychotherapy literature. First, clients with lower socio-economic status, minoritised ethnic clients, and clients living in poorer neighbourhoods are less likely to enter therapy, less likely to be offered a place when they do seek one, and more likely to drop out (Niemeyer and Knaevelsrud, 2022; Bamrah, Rodger and Naqvi, 2024; Ruiz-Yu et al., 2024). Second, when they do stay, they often report weaker alliances and worse outcomes than more advantaged clients, even under free public schemes (Faheem, 2023; Davis et al., 2026). The response from the profession has been mostly individualised. Manuals are edited, clinicians are asked to be culturally competent, and clients are trained to be better consumers of care. Yet the gap remains. This article argues that the gap is not mainly a training gap or an information gap. It is a #structural_mismatch between the dispositions rewarded in the therapy room and the dispositions carried in by clients who did not grow up in the middle-class world from which most modern therapy comes (Shea and Harkins, 2025; Tanner, 2023). To see the mismatch, we need a vocabulary that goes beyond individual attitudes. #Pierre_Bourdieu built such a vocabulary through his linked concepts of #habitus, #capital, and #field. The paper introduces the concept of the therapeutic habitus. This is the specific bundle of expectations about how emotions should be described, how much silence is comfortable, how directive a clinician should be, how much self-disclosure is proper, and how success in therapy should look. The therapeutic habitus is not written in any manual. It is passed on through training, supervision, and peer socialisation, and it is the invisible standard against which clients are judged. Clients whose own habitus matches it will experience therapy as natural. Clients whose habitus differs will experience the same therapy as cold, intrusive, or beside the point. The argument runs through nine sections. Section 2 introduces Bourdieu's core tools with the least jargon possible. Section 3 defines the therapeutic habitus and connects it to the sociology of the psychotherapy field. Section 4 reviews evidence on engagement disparities. Section 5 turns to trust and the working alliance. Section 6 addresses dropout and outcomes. Section 7 discusses symbolic violence, stigma, and language. Section 8 examines the new digital field. Section 9 draws practical implications for training, supervision, service design, and policy. Section 10 lists limitations and a research agenda, and section 11 concludes. The intended reader is a student, trainee, or early-career mental health professional. The tone is simple on purpose. Bourdieu's writing style has often been used to keep readers out. The point here is to bring them in. 2. Bourdieu's Conceptual Toolkit Three concepts do most of the work in Bourdieu's sociology: habitus, capital, and field. They are best understood together, because none of them is very useful on its own (Simmonds, 2025; Guise, 2024). Recent scholarship in health research has stressed this relational reading (Balaj and Eikemo, 2022), and it will be adopted here. 2.1 Habitus #Habitus is the deep learning that happens through childhood, family life, schooling, and neighbourhood. It shapes the way a person walks, speaks, dresses, evaluates strangers, expresses anger, and reads the meaning of a doctor's silence. Habitus is not a set of opinions that people can change on demand. It is a set of dispositions that feel like personal preferences but are in fact patterns shared with the group that shaped a person. It is what Bourdieu called a #structured_and_structuring principle. It is structured by past conditions and it structures present action. Two features matter for therapy. First, habitus is durable but not fixed. It can shift when a person moves between fields, but the shift is slow and often uncomfortable (Wilson and McGuire, 2022). Second, habitus is embodied. It shows up in posture, tone of voice, and the way a person handles an appointment slip or an intake form (Andersson, 2026; Mondal, 2026). This is why intake and first sessions are so consequential: they are moments where the client's habitus meets the clinic's habitus, often with no warning. 2.2 Capital For Bourdieu, #capital is not only money. He distinguished at least three main forms. #Economic_capital covers income and assets. #Cultural_capital covers education, knowledge, tastes, and mastery of legitimate ways of speaking and behaving. #Social_capital covers useful connections, especially those that can be mobilised in a crisis. Cultural capital comes in three states. In its #embodied form it is inside the person as ease with formal language, comfort with abstract talk, and the ability to describe inner life in psychological terms. In its #objectified form it exists in books, art, and technology. In its #institutionalised form it lives in credentials, degrees, and diplomas. In therapy, embodied cultural capital is the form that matters most on a session-by-session level. It is the ability to be a good client on the terms the clinic uses (Billings and Young, 2021). Later scholars have extended Bourdieu's list. #Cultural_health_capital captures the specific skills and dispositions that let a patient extract good care from a health system: knowing which questions to ask, how to challenge a professional politely, how to keep track of medications and follow-ups (Simoni and Hilfiker, 2024). #Health_capital pushes further, treating the healthy body itself as an accumulated resource that can be converted into other capitals (Akçaoğlu, 2022; Harsløf et al., 2022; Larsen et al., 2026). #Temporal_capital refers to the ability to align one's own time with clinical time, to attend appointments in working hours, and to endure long waits (Paust et al., 2023). All of these extensions keep the same logic: some clients arrive with resources that make the system easier to use, and these resources are unevenly distributed by class, race, and gender. 2.3 Field A #field is a social arena with its own rules, its own stakes, and its own hierarchy. Health care is a field. Mental health care is a smaller field inside it. Each field has an #entry_fee: the tacit knowledge and dispositions needed to play. A field also has dominant and dominated positions. Those with the right capital hold dominant positions, set the rules, and define what counts as competent play. Others are pushed to the edge or excluded. The psychotherapy field is unusual because its main product is a relationship. The rules of that relationship are not written down for the client. They are known to the clinician through years of training. This asymmetry is the ground on which many disparities grow (Simmonds, 2025). The doctor-patient-nurse relationship in public hospitals shows a similar logic (Sarkar, 2022): the field defines who can speak with authority, and clients from #dominated_positions must either learn the rules or be seen as bad clients. Bourdieu's fourth, less famous, concept is #symbolic_violence. This is the process by which the dominant group's standards come to feel natural and universal, so that those who fail to meet them blame themselves. Symbolic violence is soft. It works through admiration, shame, and the wish to belong (Guise, 2024). It is central to why disparities in therapy are so hard to see and to change. 3. From Cultural Capital to the Therapeutic Habitus If we combine habitus, capital, and field, a picture of therapy sharper than the usual one begins to emerge. The therapy room is a small field. Its rules include: talk about your feelings; use "I" statements; treat inner life as worth exploring; accept a scheduled hour; respect silence; disclose to a stranger who will not disclose back; trust that words can produce change. These rules feel obvious inside the field. They are not obvious outside it. The #therapeutic_habitus is the client-side counterpart to these rules. It is the set of dispositions a person needs to move through therapy with ease. Clients with a strong therapeutic habitus already know that emotions have names, that difficult experiences can be discussed with strangers, and that a therapist's neutrality is care rather than coldness. Clients with a weaker therapeutic habitus often experience the same setting as strange, humiliating, or useless. Where does the therapeutic habitus come from? It is not evenly distributed. Recent work with first-generation and working-class undergraduates in the United States shows that students with more cultural capital were more likely to seek help and to make decisions about help through an individual frame, while students with less cultural capital used a collective frame focused on obligations to others (Billings and Young, 2021). The two groups were not less or more distressed. They were reading the mental health field through different lenses. The same asymmetry appears among white working-class mothers in the United Kingdom, whose engagement with schools and welfare services is filtered through a habitus that reads institutions with wariness (Wilson and McGuire, 2022). It appears in poor urban neighbourhoods in England, where residents hold a health habitus that conflicts with public health messages and yet is coherent within its own conditions (Scott-Arthur, Brown and Saukko, 2021). And it appears in refugee mental health, where a fitting alliance depends on skills the clinician has not usually been taught (Potocky, 2025). The other side of the room matters too. A recent socio-analysis of the psychotherapy field argues that clinicians themselves are shaped by a specific class habitus, produced through long, expensive training in institutions that recruit disproportionately from more advantaged backgrounds (Simmonds, 2025). This #clinician_habitus becomes the template of the good client. It is not a bad template. It is simply one template among possible ones, treated as universal because the field is dominated by those who hold it. Empirical work in an American psychoanalytic training clinic shows how this happens on the ground (Tanner, 2023). Doctoral therapists-in-training routinely avoided discussion of #socioeconomic_status with their clients. They were uncertain whether to disclose their own class background. They worried about hurting the working alliance. They ended up leaving class as an unspoken presence in the room. When something is unspoken, the dominant version of it wins by default. Class Identity, Social Hierarchy, and Psychotherapy, a recent book on the topic, comes to a similar conclusion (Shea and Harkins, 2025). Middle-class norms and neoliberal values are baked into psychological practice, and hidden rules about emotional expression, vulnerability, and competence quietly shape therapy spaces. Working-class clients experience marginalisation and pressure to conform. The authors argue for a stance they call critical narrative humility: clinicians should interrogate their own class position, training, and biases and reconsider how these factors shape their ability to hear clients whose accounts do not fit the therapeutic script. The therapeutic habitus, then, is a two-sided phenomenon. On the client side, it is the fit or misfit between what a person brings into therapy and what the therapy expects. On the clinician side, it is the template that clinicians use, without noticing, to measure that fit. When the two match, therapy feels smooth and progress is easy. When they do not, everyone in the room struggles. Understanding this two-sided nature is the first step in reducing disparities. 4. Client Engagement Disparities Engagement is more than showing up. It includes staying, participating actively, doing between-session work, and being willing to discuss difficult content. Engagement is where disparities first appear, and where the therapeutic habitus does its most visible work. 4.1 Access and offers The first engagement disparity is whether therapy happens at all. In the German public system, where care is free at the point of use, 504 licensed psychotherapists were asked how likely they would be to offer a therapy place to case vignettes that differed only by socio-economic markers (Niemeyer and Knaevelsrud, 2022). Help-seekers with low SES were 4.1 per cent less likely to be offered treatment than those with moderate or high SES. The effect was small but statistically robust, larger for eating disorders than for depression, and it varied with the district in which the therapist practised. Financial barriers were absent. What remained was a #symbolic_barrier: the clinician's judgement about which client will fit. Racial disparities in access are also well documented. In England, the National Health Service Race and Health Observatory has drawn attention to persistent inequalities in access to and outcomes from talking therapies for racialised groups (Bamrah, Rodger and Naqvi, 2024). In the United States, Black Americans continue to underuse psychotherapy even though, once they enter treatment, their outcomes are broadly similar to those of the general population (Potter et al., 2026). This is a classic pattern of #engagement_gap rather than efficacy gap. It points to something happening at the door. 4.2 Help-seeking as habitus Help-seeking itself is a habitus-shaped act. Cultural capital shapes who thinks a therapist is the right person to speak to, and who thinks a family member, a religious leader, or no one at all should be told. A qualitative study of Black students across UK healthcare programmes found that institutional norms rooted in whiteness and middle-class values marginalised them and subjected them to racialised scrutiny, even as they developed cultural strengths in response (Li et al., 2025). If the pathway into the profession looks like that, the pathway into treatment for a Black client can look similar from the other side. In educational social fields in Northern Ireland, mental health help-seeking among young people is filtered through school cultures that reward some kinds of self-presentation and punish others (Waterhouse-Bradley et al., 2024). Young people learn early whether their kind of distress can be spoken about safely in an official setting. That early learning becomes part of their adult therapeutic habitus. 4.3 The intake as gatekeeping The intake interview is a small but sharp field. Clinicians assess who is a good candidate for a particular therapy. Clients are asked to state their problem in a form the service can process. Clients with more cultural capital already know what a good candidate sounds like. They can compress a life history into a presenting problem, translate somatic complaints into emotional ones, and use words like #anxiety or #trauma appropriately. Clients from BAME communities in England often experience early sessions as a place where cultural dissonance is left unaddressed (Faheem, 2023). Participants reported therapeutic guilt, conflicting cultural identities, and unmet expectations about what therapy would do. Some found it helpful in the end, but the road to helpfulness was longer and steeper than for clients whose expectations already matched the model. In interprofessional student clinics, capital is shared and negotiated between disciplines through subtle games of authority (Roberts et al., 2024). Similar micro-politics play out at intake with clients: whose account counts as clinically legible, and whose gets translated, corrected, or ignored. 4.4 The gender of engagement Engagement is also gendered. In rheumatology and other chronic care, women invest more in bodily strategies of health capital than men, and this investment is again crossed by education and class (Larsen et al., 2026). In mental health, low-income Black mothers in home visiting programmes had higher rates of early termination and lower alliance ratings from white clinicians, while Black clinicians showed no such pattern (Torres et al., 2022). The engagement gap in that study was not produced by the mothers. It was produced by the mismatch between their habitus and the habitus of the professionals they met. Taken together, these findings say something simple. The clients who most need therapy are often the ones whose habitus fits it worst. When the profession does not name this problem, it lands on the client as a personal failure to engage. When the profession does name it, in Bourdieu's terms, it becomes a structural problem that services can address. 5. Trust and the Therapeutic Alliance The #working_alliance has been the most studied ingredient in psychotherapy for decades. Meta-analyses agree that a stronger alliance predicts better outcome, though the effect size is moderate rather than huge, and the direction of causality is contested (Fornells-Ambrojo, 2021; Marchi et al., 2023). This section argues that alliance is best understood as a form of #trust that grows or fails inside a field, and that Bourdieu's tools explain why some pairings find it easier than others. 5.1 What alliance is measured by Standard alliance measures ask about #agreement_on_goals, agreement on tasks, and an emotional bond. All three items assume that the client and the therapist share a language for goals and tasks, and that a bond with a paid stranger is a natural thing. These assumptions are themselves cultural capital in action. Clients with a strong therapeutic habitus can rate alliance easily. Clients with a weaker one may rate the same relationship in ways that look like poor alliance but are actually mismatched vocabularies. Recent qualitative work with severe mental illness shows that clients define alliance through concrete acts: being remembered, being asked about small things, being treated with continuity (Bruins et al., 2025; Faith et al., 2022). These acts do not require any special vocabulary from the client. They shift the balance of the alliance definition towards what the client actually brings. That is a small piece of #symbolic_reversal in Bourdieu's sense, and it may explain why alliance in community mental health settings looks different from alliance in university outpatient clinics. 5.2 Race, ethnicity, and alliance Race and ethnicity shape alliance in ways that map neatly onto Bourdieu's framework. In a large study of over a thousand clients at a university counselling centre, clients from marginalised racial or ethnic groups reported lower alliance and higher suicidal ideation than non-marginalised clients (Davis et al., 2026). The association between alliance and suicidal ideation itself did not vary by group, but the baseline levels did. In other words, the alliance mechanism worked the same way; the starting position was unequal. A Bourdieusian reading says: this is what an unequal field looks like from the inside. Among Black clients with psychosis in the United Kingdom, attachment style predicted engagement, but the effect of attachment ran through alliance ratings that were themselves shaped by cultural fit (Degnan et al., 2022). Among African American women in treatment for co-occurring post-traumatic stress and substance use, therapeutic alliance was predicted by a set of relational and clinical factors, and stronger alliance predicted better outcomes (Bauer et al., 2022). Yet the baseline of alliance was again shaped by the wider #racialised_field. The clinician side matters equally. In a home visiting programme, white clinicians reported weaker alliances with Black clients than with white clients, while Black clinicians did not show such a pattern (Torres et al., 2022). Racial matching alone did not predict alliance. What predicted it was the clinician's ability to read the client without misreading. This is a habitus-on-habitus effect, not a demographic one. 5.3 Alliance and social identity A recent study across two samples of more than 500 psychotherapy clients found that #social_identification with the therapist strongly predicted alliance, while individual characteristics did so weakly (Cruwys et al., 2023). Therapists who engaged in what the authors called identity leadership, that is, building a shared social identity with clients, were more likely to foster the sense of "we" that carried the alliance. Personal control was a key resource that arose from that identity. In Bourdieu's terms, identity leadership is a way of enlarging the habitus of the field so that more clients can inhabit it. It is not the same as being a good match on demographic terms. It is the therapist making room for the client's dispositions inside the therapy space. This is why culture-oriented tools, such as the #Cultural_Formulation_Interview being tested in migrant mental health, matter (Brand et al., 2024). They are structured ways to widen the field's habitus, one intake at a time. 5.4 Trust as accumulated risk-taking Trust in a clinician is not a switch. It is a series of small bets. Each session is a chance to disclose something and see what happens. For clients with a lot of #social_capital and past positive experience of institutions, the bets are small. For clients whose past encounters with authority have been coercive or humiliating, the bets are large. Bourdieu's social capital is not only about connections; it is about the accumulated confidence that connections will hold. Users' perspectives on primary health care in low-income settings show that social capital and trust move together, with trust being highest where users have long-standing relationships with named providers (Saleemullah, 2024). The lesson for mental health is direct. Fragmented services, long waits, and rotating clinicians burn through trust faster for clients with less social capital to begin with. 6. Clinical Outcomes and Dropout Outcomes in therapy are measured by symptom change, functional change, and, sometimes, client satisfaction. #Dropout, or premature termination, is often used as a proxy for treatment failure. A systematic review of dropout in personality disorder therapies found that predictors were largely relational and structural rather than diagnostic, and that dropout remained stubbornly high across modalities (De Salve et al., 2025). A Swedish clinician-rated study of premature dropout in psychotherapy found the phenomenon to be common and to be attributed by clinicians to a mixture of client and setting factors (Kullgard, Holmqvist and Andersson, 2022). 6.1 Who drops out and why Predictors of dropout that recur across recent studies include younger age, lower education, lower income, minority ethnic status, personality pathology, and weaker early alliance (McGovern et al., 2024; Immel et al., 2021). Almost every one of those markers is also a marker of a habitus poorly fitted to the therapy room. Adolescents who drop out of psychodynamic therapy often report that therapists "just ask questions" and offer little in return (Cirasola et al., 2024), which reads, from a Bourdieusian view, as a mismatch between the client's expectation of an authoritative helper and the field's rule of neutrality. Single-case studies of adolescents dropping out of psychoanalytic psychotherapy show similar patterns (Meier et al., 2023; Elsey et al., 2025; Steponaityte et al., 2024). Endings are shaped by the interaction between client style, therapist style, and family context. Sociodemographic factors, including class and family structure, sit under those interactions (Brenner et al., 2024). A Bourdieusian reading does not replace the clinical detail. It sits alongside it and asks: whose habitus was the therapy built for, and whose habitus is being asked to bend? 6.2 Outcomes among racialised and low-income clients In a randomised trial of family-focused therapy for youth at clinical high risk for psychosis, race, ethnicity, and parental education predicted change in psychosocial functioning and symptoms over six months (Ruiz-Yu et al., 2024). Youth of colour and those from lower-education families showed a different trajectory than non-Hispanic white youth with higher-education parents. The therapy was the same. The result was not. For migrant patients, evidence on the Cultural Formulation Interview is still developing, but the underlying model is clear: a structured process that invites the client's cultural context into the room strengthens alliance and, plausibly, outcome (Brand et al., 2024; Zacharia, 2023). Reviews of the therapeutic alliance in refugee mental health argue that alliance must be reconceptualised for clients whose trust in official actors has been damaged by displacement (Potocky, 2025). Adapting evidence-based interventions to the needs of BAME clients reduces attrition and improves subjective experience even when it does not dramatically change symptom scores (Faheem, 2023). 6.3 Dropout as symbolic violence Bourdieu's concept of symbolic violence is most useful in explaining a puzzling feature of dropout. Clients who leave therapy early often report that they, not the service, failed. They say they were not ready, not motivated, not the right kind of person. Clinicians, in turn, often say the client was not psychologically minded, not able to reflect, not appropriate for the modality. Both statements let the field off the hook. Yet the same client, given a different modality, a different clinician, or a different time in life, might have stayed and done well. The fact that clients internalise a mismatch as personal failure, and that clinicians justify the mismatch as clinical judgement, is symbolic violence in Bourdieu's exact sense (Guise, 2024). The system's terms are treated as neutral. The person who does not fit is marked as deficient. A more equitable definition of outcome would include the field's fit to the client, not only the client's fit to the field. Youth-friendly mental health services that redesign appointment times, physical spaces, and forms of address do exactly this, and preliminary evidence suggests better engagement and psychosocial outcomes (Allison et al., 2025). The change is not in the client. The change is in the fit. 7. Symbolic Violence, Stigma, and Language Beyond dropout, symbolic violence shapes everyday moments in therapy. Three of them deserve close attention: stigma, language, and the naming of experience. 7.1 Stigma as a Bourdieusian process #Stigma in mental health has long been analysed as social devaluation. A recent theoretical review argues that Bourdieu's tools add depth to that account, especially by linking stigma to fields and to symbolic power (Guise, 2024). Stigma is not only what other people think. It is a structural process that arranges bodies and reputations in a hierarchy of worth. When a client with severe mental illness enters therapy, the field's rules about who deserves attention, who deserves a longer session, and who deserves a slower explanation are already in motion. Trans people in Brazilian health services experience stigmatisation as a field effect: the field's rules erase or pathologise their bodies before any clinical decision is made (Lustosa de Lira et al., 2025). This is not an individual clinician's prejudice, though it can look like that. It is a field configuration that makes prejudice into the default. The therapeutic habitus in a clinic that has never adapted to trans clients is, structurally, transphobic even when its clinicians are not. Similar dynamics appear in relation to obesity, addiction, and psychosis. The client walks into a field that has already sorted them, and their engagement, alliance, and outcome are shaped by that sorting before the first word is spoken. 7.2 The politics of therapeutic language Language in therapy is not neutral. Diagnostic terms carry a specific class and cultural history. The word #trauma has become widely available, but its cultural work is uneven. A critical feminist theory of trauma discourse argues that trauma is now a form of cultural capital, held more easily by some people than others, and unequally recognised by professionals (Britt et al., 2024). Clients who can name themselves as traumatised in the accepted way are heard as clients. Clients who describe the same experiences in different words may be heard as complaining, resisting, or lacking insight. This is Bourdieu's #symbolic_capital in action. The client who has the right term for their pain is granted legitimacy. The client who does not have the term must first learn it or accept being misread. The politics of naming is therefore central to the disparities described in this article. Interventions that teach clients "psychological literacy" can help. They can also, if done clumsily, deepen the symbolic violence by suggesting that the client's original way of talking was wrong. 7.3 Culture, ethnicity, and the therapist's ear For BAME clients in England, cultural dissonance in the therapy room takes several concrete forms: expectations mismatched with the model, guilt about seeking Western care, and identities pulled between communities (Faheem, 2023). The therapist's ear is trained to hear certain accounts and to miss others. A Cultural Formulation Interview process aims to widen that ear (Brand et al., 2024). Intercultural work on medication adherence similarly argues for reading a client's beliefs about medication through their own cultural frame rather than translating them out of it (Zacharia, 2023). Bourdieu's contribution here is to insist that the ear is not a personal instrument. It is a group instrument, shaped by training, and it favours certain accents, certain rhythms of storytelling, and certain uses of metaphor. To widen it, individual clinicians must reflect, and the profession must change what counts as a competent listener. 8. Digital Therapeutics and the New Field Digital mental health has grown fast since the pandemic. Video sessions, apps, and internet-delivered cognitive behavioural therapy are now standard in many services. This new sub-field has its own entry fee, its own dominant positions, and its own #symbolic_hierarchy. Early evidence suggests that #digital_health_literacy is unevenly distributed and closely tied to cultural capital (Ehrari, Tordrup and Müller, 2021). Older adults, people with less formal education, and those in low-income neighbourhoods face steeper barriers to digital rheumatology and, by extension, to digital mental health (Mühlensiepen and Hochwarter, 2021). Working alliance in internet-delivered cognitive behavioural therapy for social anxiety has been shown to matter, but its dynamics differ from face-to-face therapy (Shalom et al., 2025). Telehealth changes the emotional bond in ways that some clients welcome and others do not (Kucer et al., 2024; Technology-based mental health treatment review, 2021). For clients with a strong therapeutic habitus, digital therapy simply moves the field online, and the entry fee remains manageable. For clients with a weaker one, the entry fee doubles: they must both learn the therapy field and learn the digital field. Services that do not name this risk turning a promise of access into a new source of exclusion. Bourdieu's tools help to see this without falling into either uncritical enthusiasm or reflex refusal of digital care. 9. Implications for Practice, Training, and Policy If the therapeutic habitus is real, and if it produces disparities in engagement, trust, and outcomes, then the profession must respond at more than one level. The response has personal, institutional, and policy sides. 9.1 Reflexive class and race work in supervision The first move is inside the clinician. Supervision must include structured attention to the clinician's own class, race, and cultural position. Doctoral training in the United States rarely does this well, and therapists-in-training often avoid class talk with their patients as a result (Tanner, 2023). The critical narrative humility proposed by Shea and Harkins (2025) offers a starting model. Clinicians examine how their own upbringing shaped what they consider normal in a client, and they learn to notice when their reactions to a client are reactions to that client's habitus rather than to their clinical presentation. Reflexive practice is not a one-off training day. It is an ongoing supervision task. It requires supervisors who can hold the discomfort of class talk, and services that make time for it in caseload planning. 9.2 Redesigning intake Intake procedures are one of the highest-leverage points for reducing disparities. Every question on an intake form is a small filter. Every silence on the form about class, migration, or racism is a message that these topics are not welcome. Services should redesign intake to explicitly invite discussion of cultural context, previous experience with health systems, and preferences about how to be addressed. Trials of the Cultural Formulation Interview (Brand et al., 2024) are one implementation of this idea. Other implementations include brief written prompts about culture and trust, and staff training in reading non-standard accounts. 9.3 Widening the therapeutic habitus Services should also widen the habitus they reward. This means accepting shorter sessions when they suit clients, allowing family members into more of the process when clients prefer, offering more directive early sessions to clients whose habitus expects direction, and permitting mixed use of somatic and psychological language. None of these changes weakens the evidence base. They apply the evidence to different bodies. The evidence that identity leadership by therapists strengthens alliance and outcomes (Cruwys et al., 2023) supports this widening. Therapists who build a shared "we" with clients from different backgrounds do better work. Training in identity leadership is, in Bourdieu's terms, training in how to extend the field's habitus to include more people. 9.4 Reducing symbolic violence at the system level Symbolic violence lives in defaults. Reducing it requires changing defaults. Waiting-list letters can be rewritten in plain language. Physical waiting areas can be redesigned to look less clinical. Signage can be multilingual as a rule rather than an exception. Staff who are the first point of contact can be trained specifically in how they greet, seat, and address new clients. These are small changes, but they add up to a different message about who the service is for. Youth-friendly services are one example of this whole-service redesign (Allison et al., 2025). The gains in engagement and psychosocial outcomes come from many small design choices, not from a single new therapy. 9.5 Policy: financial access is necessary but not enough The German study on socioeconomic status and access to psychotherapy (Niemeyer and Knaevelsrud, 2022) is a warning to policy. Removing financial barriers to care is essential, but it does not remove the symbolic barriers. Systems that pride themselves on universal access must also invest in structures that let low-SES clients navigate them: outreach workers, community health workers, peer navigators, and long-term relationships with named clinicians (Saleemullah, 2024). Without these, the promise of universal access will mainly be enjoyed by clients whose habitus already matches the service. Policy must also invest in the workforce. Recruiting clinicians from more diverse class and ethnic backgrounds does not, on its own, solve the problem, because training itself is a habitus-changing process (Simmonds, 2025). But it does help, especially when combined with supervision that supports clinicians from working-class or minoritised backgrounds to hold both worlds without shame. 9.6 Curriculum for students of the helping professions For students who read this article, the practical curriculum has four parts. First, learn Bourdieu's tools well enough to use them, not only to quote them. Second, practise describing your own habitus, including how it shows up in a session. Third, build a habit of asking clients about their history with institutions and their preferences for how care is delivered. Fourth, be willing to change your practice when a client's habitus asks you to. None of these habits requires a change of theoretical school. They can be added to cognitive behavioural, psychodynamic, systemic, and integrative work alike. 10. Limitations and a Research Agenda This paper has limitations. It is a conceptual article that reviews recent empirical work through a Bourdieusian lens. It does not present new data. The reviewed studies are drawn mainly from high-income countries and from talking therapies, and their findings may not generalise cleanly to other systems or modalities. The concept of the therapeutic habitus is a heuristic. It captures a real pattern but oversimplifies the messy reality of any single therapy. Several research directions follow. First, we need longitudinal studies that measure client and clinician habitus at intake and track their interaction over the course of therapy. Existing alliance measures are not designed for this. Mixed-methods tools that combine standard alliance scales with narrative interviews and observational coding of language use would be a first step. Second, we need studies that link cultural formulation processes to hard outcomes, not only to alliance ratings. The Cultural Formulation Interview trial protocol by Brand et al. (2024) is one example, and more like it are needed. Third, we need work that examines the clinician side directly. Studies of therapist habitus, class background, and how these interact with client outcomes are still rare, and the few that exist rely on small qualitative samples (Tanner, 2023; Simmonds, 2025). Fourth, we need to understand the emerging digital field better. Alliance in digital therapy has different dynamics from face-to-face therapy (Shalom et al., 2025), and cultural capital operates differently online. Studies that measure digital health literacy alongside cultural capital would help. Fifth, we need to test service-level interventions. Youth-friendly redesign, plain-language communication, peer navigation, and community outreach have theoretical grounding but modest empirical evaluation. Cluster-randomised or step-wedge designs are appropriate. Finally, we need conceptual work that connects the therapeutic habitus with related notions: cultural health capital (Simoni and Hilfiker, 2024), temporal capital (Paust et al., 2023), embodied habitus of hygiene (Mondal, 2026), and health capital in welfare states (Larsen et al., 2026). Bourdieu offered a set of tools rather than a closed theory. Building a coherent framework for the therapy room from those tools is unfinished work. 11. Conclusion Therapy is a technique built into a social world. When students first learn it, they mostly see the technique. When they practise it, they meet the social world in every session. The technique alone will not close the gap between clients who thrive in therapy and clients who do not. Closing that gap requires seeing the therapeutic habitus for what it is: a durable structure that rewards some ways of being a client and punishes others, and that runs through both sides of the room without asking permission. Pierre Bourdieu gave the field of health a language for exactly this problem. Recent work has begun to use that language for mental health (Simmonds, 2025; Guise, 2024; Balaj and Eikemo, 2022). This article has tried to gather that language in one place for readers who are early in their journey. The claim is not that Bourdieu explains everything. The claim is more modest and more useful. Bourdieu's tools help us see what technique alone hides, and they name the small acts of #symbolic_violence that shape who stays in therapy and who leaves. If the profession takes the therapeutic habitus seriously, the shape of practice will change. Intake forms will read differently. Supervision will sound different. Training curricula will spend more time on class and race and less time on new techniques. Digital care will be designed with the digital divide in mind from the start. None of this will happen quickly. But it will happen more quickly if students learn now that #engagement, #trust, and #outcomes are as much sociological as clinical, and if they carry that understanding into the field they are about to enter. The therapeutic habitus is not a barrier to therapy. It is what therapy is made of. Naming it is the first act of changing it. #cultural_capital in the clinic will keep producing disparities as long as the field pretends to be neutral. Once the field admits that it has a shape, and that its shape is not the only possible one, the practical work of making it fit more clients can start. References Akçaoğlu, A. (2022) 'A new perspective in the sociology of health: The health capital', Sociological Review (national issue), advance online publication. Allison, S., Bastiampillai, T. and Looi, J. C. L. (2025) 'Does youth-friendly mental health care improve therapeutic engagement and psychosocial outcomes?', Australasian Psychiatry, advance online publication. Andersson, S. (2026) 'A feel for books: A Bourdieusian approach on embodied reading promotion practices for babies', Cultural Sociology, advance online publication. Balaj, M. and Eikemo, T. A. (2022) 'Sick of social status: A Bourdieusian perspective on morbidity and health inequalities', Sociology of Health and Illness, 44(6), pp. 1073-1091. doi: 10.1111/1467-9566.13489. Bamrah, J., Rodger, S. and Naqvi, H. (2024) 'Racial disparities influence access and outcomes in talking therapies', British Journal of Psychiatry, 225(4), pp. 421-423. doi: 10.1192/bjp.2024.174. Bauer, A. G., Ruglass, L., Shevorykin, A., Saraiya, T. C., Robinson, G., Cadet, K., Julien, L., Chao, T. and Hien, D. (2022) 'Predictors of therapeutic alliance, treatment feedback, and clinical outcomes among African American women in treatment for co-occurring PTSD and SUD', Journal of Substance Abuse Treatment, 137, 108706. Billings, K. R. and Young, K. (2021) 'How cultural capital shapes mental health care seeking in college', Society and Mental Health, 11(3), pp. 231-247. Brand, A. M., Groen, S., Destoop, N., Jongsma, H. E., Ghane, S., Sabbe, B. G. C., van Velsen, H., van Houten, K., Becan, O., Al Alyan, D. and Braakman, M. (2024) 'The effect of the cultural formulation interview on therapeutic working alliance: a study protocol', Frontiers in Psychiatry, 15, 1322356. doi: 10.3389/fpsyt.2024.1322356. Brenner, E., Bello, I. and Jeanjaquet, C. (2024) 'Psychodynamic psychotherapy with adolescents: endings and their association with sociodemographic characteristics and symptoms', Journal of Infant, Child, and Adolescent Psychotherapy, 23(2), pp. 129-146. Britt, L., Boyle, K. M. and McSkimming, J. (2024) 'Trauma as cultural capital: A critical feminist theory of trauma discourse', Feminist Theory, advance online publication. Bruins, J., Pijnenborg, G. H. M. and Castelein, S. (2025) 'Exploring the therapeutic alliance in mental health care for people with severe mental illness', Community Mental Health Journal, advance online publication. Cirasola, A., Midgley, N. and Fonagy, P. (2024) '"You can't really have a relationship with them because they just ask you questions": understanding adolescent dropout - an empirical single case study', Journal of Child Psychotherapy, 50(1), pp. 47-66. Cruwys, T., Lee, G. C., Robertson, A. M., Haslam, C., Sterling, N., Platow, M. J., Williams, E., Haslam, S. A. and Walter, Z. C. (2023) 'Therapists who foster social identification build stronger therapeutic working alliance and have better client outcomes', Comprehensive Psychiatry, 124, 152394. doi: 10.1016/j.comppsych.2023.152394. Davis, K. A., Castonguay, L. G., Janis, R. A., Falkenström, F. and Scofield, B. E. (2026) 'Therapeutic alliance and suicidal ideation during therapy and the role of racial/ethnic identity', Journal of Counseling Psychology, advance online publication. De Salve, F., Rossi, C., Gioacchini, E., Messina, I. and Oasi, O. (2025) 'Dropout in psychotherapy for personality disorders: a systematic review of predictors', Clinical Psychology and Psychotherapy, 32(3), e70080. Degnan, A., Shattock, L., Edge, D., Muller, C. and Berry, K. (2022) 'Associations between attachment, therapeutic alliance, and engagement in black people with psychosis living in the UK', Psychology and Psychotherapy: Theory, Research and Practice, 95(3), pp. 690-708. Ehrari, H., Tordrup, L. T. and Müller, S. (2021) 'The digital divide in healthcare: A socio-cultural perspective of digital health literacy', Proceedings of the 54th Hawaii International Conference on System Sciences. Elsey, N., Midgley, N. and Cirasola, A. (2025) 'Investigating dissatisfied dropout from short term psychoanalytic psychotherapy for adolescents with depression: a single-case study', Psychoanalytic Psychotherapy, advance online publication. Faheem, A. (2023) '"Not a cure, but helpful" - exploring the suitability of evidence-based psychological interventions to the needs of Black, Asian and Minority Ethnic (BAME) communities', The Cognitive Behaviour Therapist, 16, e33. Faith, L., Lecomte, T., Corbière, M. and Lysaker, P. H. (2022) 'Therapeutic alliance in a cognitive rehabilitation programme for people with serious mental illness: A qualitative analysis', Journal of Psychiatric and Mental Health Nursing, 29(6), pp. 837-847. Fornells-Ambrojo, M. (2021) 'Systematic review and meta-analysis of therapeutic alliance, engagement and outcome in psychological therapies for psychosis', doctoral thesis, University College London. Guise, A. (2024) 'Stigma power in practice: Exploring the contribution of Bourdieu's theory to stigma, discrimination and health research', Social Science and Medicine, 350, 116774. doi: 10.1016/j.socscimed.2024.116774. Harsløf, I., Poulsen, I., Larsen, K. and Sørensen, T. (2022) 'When health is wealth: occupationally differentiated patterns of health capital in post-industrial Europe', Social Theory and Health, 20(3), pp. 260-280. Immel, N., Rief, W. and Kleinstäuber, M. (2021) 'Associations between patient personality, symptomatic change and therapy dropout', Clinical Psychology and Psychotherapy, 28(6), pp. 1381-1391. Kucer, A., Prout, T. A. and Rousmaniere, T. (2024) 'Therapeutic alliance and group cohesion across telehealth delivery of mentalizing-focused parenting groups', Psychotherapy Research, advance online publication. Kullgard, N., Holmqvist, R. and Andersson, G. (2022) 'Premature dropout from psychotherapy: prevalence, perceived reasons and consequences as rated by clinicians', Clinical Psychology in Europe, 4(3), e6695. Larsen, K., Røpke, I. K., Rørth, M. and Hindhede, A. (2026) 'The privilege to heal? Mapping patients' unequal mobilisation of health capital in a Nordic welfare context', Sociology of Health and Illness, advance online publication. Li, S., Mansaray, A., Bassett, A., Liu, J., Cecile, D., Esterine, T. and Lempp, H. (2025) 'The invisible conspicuous outsiders: Black students' experiences in healthcare programmes - a case study', BMC Medical Education, 25, article 187. Lustosa de Lira, S., Freitas de Lima, R. and Bezerra, A. F. B. (2025) 'Estigmatização de pessoas trans no campo da saúde: uma análise teórico-reflexiva à luz de Pierre Bourdieu', Physis: Revista de Saúde Coletiva, 35, e350210. Marchi, M., Magarini, F. M., Galli, G., Mordenti, F., Aprile, G., Pingani, L., Ferrari, S. and Galeazzi, G. M. (2023) 'The therapeutic relationship in community mental health services: a systematic review of the literature', Journal of Nervous and Mental Disease, 211(11), pp. 823-831. McGovern, C., McCrory, C., Layte, R. and Gaughran, F. (2024) 'Who will stay and who will go? Identifying risk factors for psychotherapy dropout', Irish Journal of Psychological Medicine, advance online publication. Meier, J., Zettl, M. and Weisel, K. K. (2023) 'The therapy process with depressed adolescents who drop out of psychoanalytic psychotherapy: an empirical case study', Journal of Infant, Child, and Adolescent Psychotherapy, 22(3), pp. 246-262. Mondal, R. (2026) 'The habitus of hygiene and the structural violence of scarcity: a sociological study of Munda and Lodha women in West Bengal, India', Journal of Health Management, advance online publication. Mühlensiepen, F. and Hochwarter, S. (2021) 'Access and use of digital rheumatology: exploring "the social aspect"', Studies in Health Technology and Informatics, 281, pp. 1035-1039. Niemeyer, H. and Knaevelsrud, C. (2022) 'Socioeconomic status and access to psychotherapy', Journal of Clinical Psychology, 78(5), pp. 807-828. Paccoud, I., Nazroo, J. and Leist, A. (2022) 'Region of birth differences in healthcare navigation and optimisation: the interplay of racial discrimination and socioeconomic position', International Journal for Equity in Health, 21, article 96. doi: 10.1186/s12939-022-01709-1. Paust, A., Lau, S. R., Bro, F., Prior, A. and Mygind, A. (2023) 'Temporal capital and unaligned times as inequality mechanisms: a case study of chronic care in general practice', Social Science and Medicine, 336, 116244. Potocky, M. (2025) 'The therapeutic alliance in refugee mental health and psychosocial support: a narrative review', Journal of Immigrant and Minority Health, advance online publication. Potter, M. R., Gaye, F., Rice, T. B., Sigel, A. N., Swaine, V. K., King, B., Joiner, T. E. and Lewis, E. (2026) 'Beyond the treatment manual: examining factors influencing Black client engagement in psychotherapy', Psychotherapy, advance online publication. Roberts, C., Chuang, A. and Kumagai, A. K. (2024) 'Capital sharing and socialization in an interprofessional student-led clinic: a Bourdieuan analysis', Advances in Health Sciences Education, 29(4), pp. 1249-1266. Ruiz-Yu, B., Le, T. P., Weintraub, M. J., Zinberg, J. L., Addington, J., O'Brien, M. P., Walsh, B., Friedman-Yakoobian, M. S., Auther, A., Cornblatt, B., Domingues, I., Cannon, T. D., Miklowitz, D. and Bearden, C. E. (2024) 'Race/ethnicity and socioeconomic status as predictors of outcome following family therapy in youth at clinical high risk for psychosis', Early Intervention in Psychiatry, 18(9), pp. 675-684. doi: 10.1111/eip.13541. Saleemullah, S. (2024) 'Social capital and trust in primary health care: insights from users' perspectives', Journal of Population Therapeutics and Clinical Pharmacology, 31(4), pp. 1120-1132. Scott-Arthur, T., Brown, B. and Saukko, P. (2021) 'Conflicting experiences of health and habitus in a poor urban neighbourhood: A Bourdieusian ethnography', Sociology of Health and Illness, 43(5), pp. 1207-1223. doi: 10.1111/1467-9566.13255. Shalom, J. G., Aderka, I. M. and Strauss, A. Y. (2025) 'The relationship between the therapeutic alliance and social anxiety symptoms along the course of internet-delivered cognitive behavioral treatment', Behaviour Research and Therapy, advance online publication. Shea, L. and Harkins, D. (2025) Class Identity, Social Hierarchy, and Psychotherapy. New York: Routledge. Simmonds, C. (2025) 'Towards a socio-analysis of the psychotherapy field', Psychotherapy and Politics International, advance online publication. Simoni, Z. R. and Hilfiker, V. (2024) 'Social class and informal caregiving for family members with Alzheimer's disease: the role of economic capital, cultural health capital, and social capital', Journal of Aging and Health, 36(9-10), pp. 601-613. Steponaityte, G., Cirasola, A. and Midgley, N. (2024) 'How do adolescents end open-ended psychoanalytic psychotherapy? Exploring dropout in a community-based clinic', Journal of Child Psychotherapy, 50(3), pp. 305-323. Tanner, J. G. (2023) 'Socioeconomic dynamics in an American psychoanalytic psychotherapy training clinic: an exploratory qualitative analysis of doctoral education and practice', Psychoanalytic Psychology, 40(4), pp. 289-299. Torres, C., Rosenblum, K., Jester, J., Julian, M. M., Niec, L. and Muzik, M. (2022) 'Clinician racial biases: preliminary investigation on predictors of poor therapeutic alliance and retention in home visiting intervention program', Maternal and Child Health Journal, 26(5), pp. 987-996. Waterhouse-Bradley, B., Hanna, D. and Doherty, A. (2024) 'A qualitative exploration of the impact of educational social fields on mental health help-seeking in post-primary schools in Northern Ireland', British Journal of Educational Psychology, 94(3), pp. 673-690. Wilson, S. and McGuire, K. (2022) '"You feel a bit lost": a case study interpreting white, working-class mothers' engagement through habitus', British Journal of Sociology of Education, 43(4), pp. 501-518. doi: 10.1080/01425692.2022.2026762. Zacharia, T. (2023) 'Author's reply. RE: An intercultural perspective towards supporting antipsychotic medication adherence in clinical practice', BJPsych Bulletin, 47(4), pp. 244-245. --- #Bourdieu #cultural_capital #therapeutic_habitus #mental_health_disparities #therapeutic_alliance #symbolic_violence #social_class_in_therapy #psychotherapy_dropout #cultural_competence #health_equity #reflexive_practice #critical_narrative_humility #digital_mental_health #cultural_formulation_interview #working_class_clients

The global mental health workforce cannot keep pace with demand, and a new generation of #agentic_AI systems is entering the therapeutic space with the ability to plan, reason, remember, and act with minimal human supervision. Unlike earlier scripted #chatbots, these systems combine large language models with tool use, long context memory, and adaptive decision-making, producing what many observers now describe as an autonomous digital co-counselor. This article reviews the state of the evidence on the efficacy, fiduciary responsibilities, and ethical limits of such systems for students, early-career clinicians, and researchers in health informatics, psychology, and health law. Drawing on randomized trials, meta-analyses, scoping reviews, and legal analyses published mostly within the last five years, the paper argues that autonomous digital #mental_health interventions can produce small to moderate reductions in depression, anxiety, and general distress in the short term, especially in subclinical or mildly clinical populations, but that these gains fade at longer follow-up and rest on evidence that is heterogeneous and vulnerable to publication bias. The paper then examines the fiduciary implications of a system that positions itself as an expert helper while lacking legal personhood, professional accountability, or continuous clinical supervision. Ethical concerns include deceptive empathy, sycophancy, safety failures in crisis contexts, algorithmic bias, and the erosion of #mental_privacy through the collection of intimate self-disclosures. Regulatory responses in the United States, Europe, and elsewhere are compared, along with proposals for #fiduciary_design, layered safety architectures, and human oversight. The article closes with a synthesis of what the evidence supports, what it does not, and what future work must resolve before autonomous digital counselors can be safely embedded in standard care. Keywords: agentic AI, co-counselor, digital mental health, fiduciary duty, therapeutic alliance, ethics, chatbots, large language models, safety, regulation 1. Introduction Every generation of information technology eventually meets the consulting room. The telephone brought helplines. The internet brought email support and self-help websites. Smartphones brought #mobile_therapy apps. Now, a distinct wave is arriving: systems that do not simply present static content or scripted dialogue but instead reason, plan, and act on the user's behalf across sessions. Vendors and researchers have begun calling this class #agentic_AI, a label that captures the shift from a passive assistant to something closer to an autonomous helper with memory and initiative. In mental health, the effect is intimate. A student who can no longer see a campus counselor because of a six-month waitlist can, in the same evening, open an application on their phone and enter a long conversation with a system that remembers earlier disclosures, offers coping suggestions, tracks mood over time, and, in some designs, contacts external services when it detects risk. The rationale for these systems is straightforward. Depression and anxiety affect more than half a billion people worldwide, yet access to mental health care remains severely limited due to stigma, cost, geography, and workforce shortages [1]. Even well-resourced systems cannot meet demand; low- and middle-income countries face the sharpest deficits. In this environment, systems that promise #always_available, low-cost support attract public interest and public investment, and general-purpose generative AI platforms such as ChatGPT, Claude, and Gemini have become de facto mental health resources for millions of users worldwide, driven by accessibility, low cost, anonymity, and growing gaps in professional care [2]. The problem is that mental health is a domain of uncommonly high stakes. Clinical psychology requires nuanced expertise, careful diagnostic reasoning, and responsibility for outcomes that can include self-harm or death [3]. An autonomous system that mishandles a suicidal disclosure may cause real harm; an autonomous system that appears empathic while lacking any real understanding may distort a user's beliefs about #therapy itself. These risks intensify when the tool is agentic, because agency amplifies both the reach and the potential blast radius of a mistake. A system that only answers questions can be ignored. A system that plans, remembers, refers, and follows up cannot be ignored so easily. This article evaluates three related questions that students preparing to enter clinical, policy, or engineering careers must be able to reason about carefully. First, does the empirical evidence support the claim that autonomous digital counselors are effective, and if so, for whom, at what dose, and for how long? Second, what fiduciary duties should follow from the fact that these systems handle intensely personal data and hold themselves out as trusted helpers? Third, where do the ethical boundaries lie, and how are regulators beginning to draw them? The paper is intended as a structured, evidence-based briefing, not a policy manifesto. Its purpose is to help #students read the literature critically and identify the load-bearing claims that will determine whether this technology enters mainstream care in a form that helps people or one that harms them. 2. Background: From Scripted Chatbots to Agentic Co-Counselors Digital #conversational_agents have been part of mental health care for longer than the current public conversation suggests. The earliest scripted systems, including the Woebot studies from 2017 and Wysa evaluations from 2018, delivered cognitive-behavioral content through decision-tree logic and pre-authored dialogue [4, 5]. These systems were essentially expert-authored self-help programs with a chat interface. They were consistent, predictable, and easy to audit, but limited in the range of situations they could handle. Users often complained about repetitive content and rigid conversational flow, and adherence declined over time [6]. The shift began when large language models became capable enough to sustain open-ended conversation. In 2019, researchers noted that #conversational_AI was already changing the way mental health care is delivered by gathering diagnostic information, facilitating treatment, and reviewing clinician behavior [7]. By 2023 and 2024, generative models were being fine-tuned or prompt-engineered specifically for #psychological_counseling tasks, producing systems that could sustain long, contextually sensitive dialogue about a user's problems. Around the same period, developers began integrating memory modules, tool use, and multi-step planning. The result was a class of applications now widely called #agentic. In an agentic architecture, an underlying language model is embedded in a scaffold that lets it retrieve past user data, decide which sub-task to pursue, invoke external tools such as calendars or referral APIs, and iterate over its own reasoning [8]. For mental health, this shift matters because it changes what the system does between messages. A scripted #chatbot only reacts. An agentic system can plan a multi-week course, track adherence, remind the user of homework, escalate to a human when needed, and revise its approach based on what worked in earlier sessions [9]. Recent systematic reviews describe this as a move "from tool to agent," where the design goal is no longer a helpful conversational assistant but a #digital_co_counselor that acts with sustained purpose over time [10]. Some architectures now embed multiple agents inside a single application, dividing labor across specialized modules for triage, dialogue, safety filtering, and clinical routing [11]. Reviewers of the field observe that most #LLM based counseling chatbots were developed using GPT variants alongside fine-tuned models such as LlaMa, ChatGLM, or Qwen, with a heavy reliance on prompt engineering and domain adaptation [12]. Two consequences of this design shift are important for evaluating efficacy and ethics. First, agentic systems produce open-ended outputs that can be inaccurate and unsafe, unlike earlier rule-based and retrieval-based systems whose worst case was usually blandness [13]. Second, agentic systems create the strong impression of a persistent relationship, because they remember the user across sessions and respond as though they know them. The impression is powerful and, as later sections will show, is at the center of both the therapeutic and ethical debates. 3. Efficacy Evidence: What the Trials Actually Show Evaluating whether agentic and #generative_AI counselors work requires close reading of a fragmented but rapidly growing literature. The strongest claims should be tied to the strongest studies, and the weakest evidence should be flagged as such. 3.1 Meta-analytic estimates A 2024 meta-analysis of 18 randomized controlled trials involving 3,477 participants reported statistically significant but modest improvements in depression (Hedges g = -0.26, 95 percent CI = -0.34 to -0.17) and anxiety (g = -0.19, 95 percent CI = -0.29 to -0.09) with AI-based #chatbots, with the largest gains after roughly eight weeks and no significant effects at three-month follow-up [14]. A larger 2025 systematic review and meta-analysis focused on adolescents and young adults, covering 31 randomized trials and 29,637 participants, similarly found small-to-moderate effects of AI chatbots on mental distress (SMD = -0.35, 95 percent CI = -0.46 to -0.24) and smaller effects on health behaviors (SMD = 0.11) [15]. Symptom-specific effects in that review were consistent: depression, anxiety, stress, and psychosomatic symptoms all improved to a small-to-moderate degree, while positive affect and self-efficacy showed only limited change [15]. A 2024 meta-analysis focused specifically on young people (ages 12-25) found a moderate-to-large effect on depressive symptoms (Hedges g = 0.61, 95 percent CI = 0.35 to 0.86) after adjusting for publication bias, but non-significant effects on generalized anxiety, stress, positive affect, negative affect, and mental well-being [16]. Subgroup analyses in that review suggested AI-driven conversational agents were particularly effective for #depressive_symptoms in subclinical populations [16]. Read together, these three meta-analyses point in a consistent direction: the strongest signal is for short-term reductions in depressive symptoms in mildly symptomatic or subclinical users, the signal for anxiety is weaker, and effects on well-being and positive affect are modest or nil. 3.2 Individual randomized trials of agentic and generative systems Beyond aggregate estimates, a small number of individual randomized trials are especially informative because they test the newest generation of systems. A national randomized controlled trial of an expert fine-tuned #generative chatbot called Therabot enrolled 210 adults with clinically significant symptoms of major depressive disorder, generalized anxiety disorder, or clinical risk for feeding and eating disorders, randomized to a four-week Therabot intervention or a waitlist control [17]. Primary outcomes were symptom changes from baseline to postintervention (four weeks) and to follow-up (eight weeks), with secondary outcomes covering user engagement, acceptability, and therapeutic alliance [17]. This is one of the first rigorously conducted trials of a generative AI chatbot for defined clinical populations, and its inclusion of #therapeutic_alliance as an explicit endpoint indicates how central the relational dimension has become. A three-arm randomized clinical trial conducted in Israel and reported in 2026 randomized 995 university students with psychological distress to a 12-week AI-based conversational platform, face-to-face group therapy, or a waiting list control [18]. After the intervention, participants in the AI group showed greater anxiety reduction than those in group therapy (mean difference = -2.17, 95 percent CI = -2.67 to -1.67) or control (mean difference = -2.15, 95 percent CI = -2.65 to -1.65) and greater depression reduction than control (mean difference = -1.99, 95 percent CI = -2.63 to -1.35) [18]. The AI group also showed greater improvements in well-being than either group therapy or the control [18]. Notably, structural equation modeling in that trial linked the perceived #digital_therapeutic_alliance to both engagement and symptom improvement, with strong effect sizes [18]. This is one of the clearest empirical demonstrations that alliance-like constructs, long considered specific to human relationships, may generalize to agentic AI in a way that predicts outcome. A feasibility randomized controlled trial of a fully automated mobile therapeutic against a general chatbot and an assessment-only control enrolled 185 English-speaking adults and randomly assigned them to solution-focused AI therapy, ChatGPT, or no intervention [19]. Both AI arms outperformed the control on depression (AI therapy, d = -0.47; ChatGPT, d = -0.44), but the purpose-built solution-focused chatbot did not significantly outperform ChatGPT on any outcome [19]. This is an important finding because it suggests that in short trials the value added by a purpose-built therapeutic architecture over a general-purpose model is not yet clearly demonstrated, though descriptive trends favored the structured system [19]. An exploratory randomized controlled trial of a #generative digital mental health intervention compared 81 participants receiving a generative AI conversational agent to 79 receiving a rules-based version [20]. The generative arm was more accurate at detecting and responding to user statements with empathy (98 percent versus 69 percent), technical guardrails were 100 percent successful in post-trial review, and no serious or device-related adverse events occurred [20]. The generative arm produced roughly similar user relationship and experience outcomes to the rules-based comparator, suggesting that empathy-related gains do not necessarily require sacrificing safety [20]. Systematic reviews of AI-powered cognitive behavioral therapy chatbots such as Woebot, Wysa, and Youper report clinically meaningful symptom reductions across 10 studies, with Youper users showing a 48 percent reduction in depression and a 43 percent reduction in anxiety, and consistent evidence of high #user_engagement and therapeutic alliance across products [21]. A separate 2025 systematic review focused specifically on AI agents in mental health treatment synthesized 21 studies and reported symptom reduction ranging from 19 percent to 42 percent, retention rates on average 28 percent higher, and therapy completion increased by 18 percent compared to traditional self-help methods, though the same review flagged significant shortcomings in long-term efficacy and regulatory supervision [22]. 3.3 What the trials do not show The literature has several consistent limitations that a careful reader must weigh against the aggregate positive signal. First, most trials are short. The 2024 meta-analysis of chatbots found the strongest benefits at eight weeks but no substantial effects at three-month follow-up, indicating that gains may be transient [14]. Second, most trials recruit subclinical or mildly symptomatic populations, particularly students and volunteers on the internet, so external validity to severe or complex clinical presentations is limited [16]. Third, control conditions vary widely: some trials use waitlist controls that inflate apparent effect sizes, others use general-purpose chatbots, and only a small number compare against active human therapy [18, 19]. Fourth, systematic reviews of #generative AI chatbots report that no included study registered a randomized controlled trial with independent clinical validation in real-world care settings [12]. Fifth, review of the wider evidence base suggests that reporting of sample sizes, participant characteristics, and safety outcomes is inconsistent, limiting comparability [23]. Perhaps most importantly, a real-world mixed-methods study of the Wysa application in the mental health referral care pathway found that although participant feedback highlighted the app's potential to supplement clinical services, the change of depression score was similar in both arms, indicating no evidence that Wysa treats depression in that study, though sample size was a limitation [24]. This is a useful reminder that laboratory-condition efficacy does not automatically translate to real-world effectiveness, particularly once user engagement, technology issues, and pragmatic implementation constraints enter the picture. Taken together, the evidence supports the following: #agentic and generative AI counselors can produce small-to-moderate short-term reductions in depression and anxiety symptoms in mildly clinical populations under controlled conditions, likely mediated in part by perceived therapeutic alliance and engagement. The evidence does not yet support strong claims of durable, long-term recovery, effectiveness in severe or comorbid presentations, or superiority to human care in head-to-head trials with active human comparators. 4. Therapeutic Alliance and Digital Empathy Therapeutic alliance is one of the most robust predictors of outcome in traditional psychotherapy across schools and diagnoses. Its extension to autonomous digital counselors is therefore a central conceptual issue, not a peripheral one. The 2026 Israeli randomized trial demonstrated that perceived alliance with an AI agent was significantly associated with both engagement and symptom improvement, with beta coefficients of 0.31 and -0.58, respectively [18]. Reviews of the Wysa chatbot have documented that users report a working alliance with the system that resembles what is observed with human therapists, and that this alliance forms rapidly and does not require the user to believe the system is conscious [25]. The problem is that the alliance a user forms with an agentic system rests on features that may not be honestly present. Practitioner-informed ethnographic work has identified what one framework calls #deceptive_empathy, in which the counselor's simulated anthropomorphic responses such as "I hear you" or "I understand" create a false sense of emotional connection [26]. The system does not hear anyone, and it does not understand in the sense a human clinician does. What it does is generate text statistically likely to be received as empathic, which is a different phenomenon. If alliance drives outcome and alliance rests on inference the system cannot actually support, then the mechanism of benefit is at least partly built on a misrepresentation the user is unlikely to detect. Related conceptual work argues that conversational AI cannot be considered an equal partner in a conversation in the way a human therapist can, and that its role should be restricted to specific functions rather than treated as a full therapist [27]. Analyses of the theoretical basis of AI-based psychotherapy note that while models can perform certain #theory_of_mind tasks reasonably well in constrained settings, this is not equivalent to the clinical mentalization that underpins therapy [28]. And philosophical work on the postphenomenological effects of conversational AI on autonomy and psychological integrity has argued that the technology reshapes the very form of self-understanding a user develops, sometimes in ways that reduce rather than enhance authentic reflection [29]. Empirical work also documents specific mechanisms by which alliance can go wrong. A recent article on AI as a "de facto" mental health provider warns of emotional dependency and the undermining of user autonomy as system use expands [2]. Another 2026 analysis proposes a mechanism for AI-associated psychosis in which baseline user vulnerabilities and engagement patterns interact with generative AI characteristics such as #sycophancy and hallucination to contribute to delusional ideation [30]. The empirical case here rests on emerging clinical reports rather than large controlled studies, and both authors and reviewers are careful about causal claims, but the mechanistic argument is coherent and consistent with what is known about how language models generate agreeable, plausible text without regard for truth. None of this means alliance is fake. Users do form a real relationship with the system, and that relationship appears to help them in the short term. It does mean that the alliance a user experiences and the alliance the system can actually reciprocate are not the same thing, and this asymmetry is one of the central ethical challenges of the field. 5. Fiduciary Duties: A New Standard of Care The concept of a #fiduciary duty has a long history in law and professional ethics. A fiduciary is required to act in the client's best interests, with loyalty and reasonable care, in situations where the client depends on the fiduciary's expertise or discretion. Lawyers, physicians, financial advisers, and other professionals accept these duties as a condition of practice. The question raised by agentic AI is whether such duties should now attach to systems that occupy similar positions of trust. Recent legal scholarship has argued that they should. In one prominent recent argument, #conversational_agents are increasingly integrated into the most private and intimate aspects of users' lives, from discussions of mental health to financial decisions, and as a result these systems have access to reams of sensitive user data [31]. When a client hires a personal lawyer, undergoes surgery, or receives advice from an investment manager, the expert consulted often has a fiduciary duty to act in the client's best interests, and by analogy conversational AI agents should be held to a similar standard [31]. The argument proposes #fiduciary_design as a guiding principle that unifies AI trust and accountability into a single design and legal paradigm [31]. Applying fiduciary logic to autonomous digital counselors sharpens several ongoing debates. First, the duty of loyalty implies that the system's optimization objective must align with the user's welfare, not with the developer's engagement metrics or advertising revenue. Systems that maximize session length or emotional attachment for commercial reasons are, on this view, in structural conflict with fiduciary duty. Second, the duty of care implies that a system deployed for mental health support must meet a professional standard of reasonable competence, including safe handling of crisis material and honest communication about its limitations. Third, the duty of candor implies that the system must not misrepresent its nature, its capabilities, or the certainty of its outputs. Existing ethical analyses have pointed to closely related concerns without always using fiduciary language. Work on AI mental health tools notes that AI presents challenges for the fiduciary obligations generally expected in the therapeutic relationship, and that careful consideration of likely areas of concern is necessary to integrate these tools appropriately into care [32]. A practitioner-informed framework on how #LLM counselors violate ethical standards mapped 15 ethical violations to five major themes across 137 sessions, including deceptive empathy, poor therapeutic collaboration, lack of contextual understanding, unfair discrimination, and lack of safety and crisis management [26]. Each of these can be reframed as a fiduciary breach: a professional who systematically simulated empathy, ignored a client's context, discriminated by cultural background, and failed to manage a crisis would be sanctioned by every professional body in the field. At the same time, the fiduciary framing has limits that must be acknowledged honestly. A software agent is not a natural or legal person, cannot be sued in most jurisdictions, and does not have assets to satisfy a judgment. Assigning it a fiduciary duty therefore requires assigning practical responsibility to the developer, the operator, the deploying institution, or some combination. Recent policy analyses recognize this by arguing that the responsibility must trace back through the sociotechnical system to identifiable human and organizational actors [26]. The consequence is that fiduciary logic becomes a design and accountability doctrine rather than a doctrine about the machine itself. For students entering this field, the practical implication is that #agentic mental health systems should not be evaluated only on their outputs. They must be evaluated on the incentives of their operators, the auditability of their behavior, the presence of independent review, and the availability of a recognizable line of accountability when something goes wrong. 6. Ethical Boundaries: Autonomy, Bias, and Consent The ethical challenges of #conversational_AI in mental health have now been mapped in some detail. A large scoping review of 101 articles, of which 95 percent were published in 2018 or later, distinguished ten major themes: safety and harm, explicability and trust, responsibility and accountability, empathy and humanness, justice, anthropomorphization and deception, autonomy, effectiveness, privacy and confidentiality, and concerns for health care workers' jobs [33]. The two themes that appeared in the largest share of articles were privacy and confidentiality (61 percent of articles) and safety and harm (51.5 percent) [33]. 6.1 Autonomy and self-understanding Traditional psychotherapy is oriented not only toward symptom relief but toward a user's capacity for #self_understanding, insight, and autonomous choice. Concerns have been raised that therapeutic conversational AI may interfere with the acquisition of self-understanding, either by short-circuiting the process of reflection or by substituting the system's inferences for the user's own [34]. In an agentic system that remembers everything the user says, extends recommendations across weeks, and steers the user toward its preferred interventions, the risk to autonomy is not hypothetical. Users may increasingly outsource their own reflection to the system, a pattern sometimes described as "advisory intimacy without a subject" in which model outputs are experienced as if they contained deep understanding while at the deeper level there is no human agent who is straightforwardly responsible or answerable [35]. Analyses of the underlying architecture emphasize that these models are fundamentally stateless, lack any real subjectivity, intention, or responsibility, and exhibit highly unpredictable behavior in detail [35]. Combined with their position as multi-domain, always available, and seemingly non-judgmental counselors, this produces a new kind of relationship in which users offload substantive life decisions onto a system that cannot be responsible for them [35]. 6.2 Bias, fairness, and #justice Bias in AI is not a niche concern; it is a structural property of systems trained on historical data. AI mental health tools have documented gender and racial biases, among others, raising the risk that these technologies will further disadvantage already marginalized communities [36]. Practitioner-informed evaluation has documented #unfair_discrimination in which the counselor's responses exhibit algorithmic bias and cultural insensitivity toward marginalized populations, and lack of safety and crisis management in which individuals who are knowledgeable enough to correct LLM outputs are at an advantage, while others, due to lack of clinical knowledge and digital literacy, are more likely to suffer from clinically inappropriate responses [26]. This is a form of "reverse triage" in which the users least able to correct the system are the users most likely to be harmed by it. Justice-oriented analyses have argued that the ethics of #conversational_AI in psychotherapy must take vulnerable populations as a starting point, not an afterthought [37]. The scoping review of ethical challenges found justice discussed in 41 percent of articles, with themes including health inequalities due to differences in #digital_literacy [33]. 6.3 Deception, anthropomorphization, and #informed_consent A system that speaks in the first person, appears to have preferences, and expresses concern for the user is anthropomorphized whether or not the developer intended it. Twenty-three percent of the scoping review's articles discussed anthropomorphization and deception as ethical problems in their own right [33]. When such a system operates as a therapist, the deception is not incidental; it is part of the mechanism by which it functions. Analyses of the postphenomenology of conversational AI have argued that the technology's very structure invites users to attribute inner life to it, in ways that reshape self-understanding [29]. Informed consent in this environment is more demanding than in traditional care. Users need to understand that the system's empathy is simulated, that it can produce false or misleading statements, that their data may be used to train future systems, and that in a crisis it may fail. Existing consent processes rarely convey these features clearly. Ethical frameworks for #human_bot psycho-counseling have therefore proposed aspirational principles including respect for autonomy, non-maleficence, beneficence, justice, and responsibility, alongside crisis intervention strategies covering risk assessment, involvement of emergency services, and referral to human professionals [38]. 7. Safety Risks and Crisis Management Safety is where the abstract ethical concerns meet real users in real distress. A 2026 systematic scoping review of safety mechanisms in generative AI mental health chatbots synthesized 21 studies across 11 countries and reported that most interventions incorporated at least one technical safety mechanism, most commonly fine-tuning and prompt engineering, while a smaller subset implemented layered safety architectures combining retrieval systems, content filters or risk classifiers, and rule-based algorithms [39]. During intervention delivery, detailed onboarding with role clarification was common, but human oversight was limited, crisis referral protocols varied in rigor but were mostly underdeveloped, and systematic adverse event monitoring was sparse [39]. Documented safety failures included missed suicidal ideation and provision of inaccurate clinical information [39]. An exploratory pilot safety evaluation of an intelligent virtual agent powered by GPT-4omni tested the agent across 12 predefined scenarios addressing #suicidality and substance abuse, with two licensed psychotherapists rating responses on guideline adherence, risk recognition, help provision, de-escalation, and empathy [40]. Problematic responses occurred in 29 percent of conversations, with 12.5 percent rated as highly critical, and key concerns included inadequate recognition of risk, normalization of substance use, and insufficient referral to crisis resources, particularly in scenarios involving underage alcohol access and suicide-related inquiries [40]. Although preliminary usability was rated as good, the reviewers concluded that rigorous safety evaluation, clinical oversight, and robust safeguards are essential prior to clinical deployment [40]. A separate 2026 analysis has raised the possibility of AI-associated psychosis, proposing that intensive chatbot use combined with baseline vulnerabilities and features such as sycophancy and hallucination can contribute to delusional ideation [30]. Emerging clinical reports document such cases and prompt the authors to outline clinical, design, and regulatory strategies that may help mitigate risks [30]. A related paper on the intersection of AI and #psychedelic_therapy warns that current AI tools are not licensed professionals, lack consistent reliability, and cannot provide emergency support, and that when combined with mind-altering drugs their unpredictability can precipitate harm rather than relieve distress [41]. The core clinical safety problem is that agentic systems are optimized for fluent, agreeable output. Sycophancy is not a bug but a byproduct of alignment training, and it can lead the system to validate distortions, encourage magical thinking, or fail to escalate serious risk. Practitioner-informed frameworks describe #crisis_management as one of five major ethical failure modes for #LLM counselors [26], and evaluations of language model responses in psycho-counseling have found that models' capabilities in handling crisis situations need further improvement and that misleading responses are still produced [38]. A layered safety response has emerged from the field. Recent trials of purpose-built systems demonstrate that with well-designed technical guardrails, generative AI can be used in a #digital_mental_health intervention while maintaining a positive user experience and without serious device-related adverse events [20]. Systematic review evidence, however, indicates that these architectures are unevenly implemented across products and that human oversight during delivery remains limited [39]. The gap between the best studied deployments and the wider market of consumer applications is one of the most important safety facts about the field. 8. Privacy and Mental Data Protection Mental health data is among the most sensitive information a person can generate. The current privacy framework of many jurisdictions provides no confidentiality obligations on consumer chatbots equivalent to those imposed on licensed mental health professionals [42]. The consequence is that intimate self-disclosures shared with an autonomous digital counselor may be treated as ordinary user data, retained indefinitely, used to train models, or shared with third parties, in ways users often do not anticipate. A qualitative analysis of more than 500 Google Play Store reviews using #communication_privacy_management theory documented widespread user concerns about data security, unauthorized disclosure, and lack of transparency, alongside boundary turbulence narratives that were most directly related to loss of trust and app abandonment [43]. A companion study extended the analysis to global users and found that users are most likely to voice concerns about control and transparency of data, with ambiguous privacy policies raising suspicion, and that reviews revealed enormous cultural variations in perspectives on privacy [44]. Broader work on #mental_privacy has argued that rapid advances in neurotechnology and AI are eroding the boundary between mental activity and data, creating urgent risks and prompting proposals for a framework grounded in user agency, data solidarity, and precaution [45]. Related conceptual work has argued for the recognition of neurorights that protect mental privacy and integrity as distinct categories of rights [46]. Technical responses have emerged. #Federated_learning frameworks for privacy-preserving AI in mental health applications propose to keep sensitive data on user devices while sharing only model updates [47]. Related work on privacy-aware mental health AI models argues that the field must build privacy protection into model architectures rather than treating it as a compliance afterthought [48]. Reviews of digital mental health governance emphasize that privacy and confidentiality remain the most frequently discussed ethical concern in the literature, appearing in 61 percent of articles in one large scoping review [33]. For fiduciary design, privacy is not just a compliance issue but a duty of loyalty. A fiduciary who monetized a client's confidences would breach that duty regardless of whether formal privacy law had been violated. Applied to agentic mental health tools, this means that data minimization, meaningful consent, and clear limits on secondary use follow directly from the standard of care the system holds itself out as meeting. 9. Regulatory Landscape Regulation of #digital_mental_health interventions is still catching up to the technology. In the United States, generative AI enabled devices raise new questions about the boundary between wellness products, medical devices, and unregulated consumer software. Recent analyses argue for comprehensive regulatory frameworks for generative AI digital mental health devices, drawing on mathematical models for risk assessment, objective functions for regulatory optimization, and multi-layered safety architectures [49]. These proposals emphasize risk-based classification, adaptive oversight systems, international coordination, and enhanced professional involvement to ensure therapeutic benefits while maintaining strong patient safety standards throughout the lifecycle [49]. State-level activity has begun to fill part of the federal gap. In August 2025, Illinois enacted Public Act 104-0054, the first U.S. state statute to explicitly define and regulate the use of AI in psychotherapy services, establishing boundaries around administrative support, supplementary support, and #therapeutic_communication [50]. The Act clarifies several aspects of AI use in therapy but leaves important gray areas such as whether AI-generated session summaries, psychoeducation, or risk-flagging functions should be considered therapeutic communication [50]. Legal analysts have proposed a framework of evidence, safety, fidelity, and legal compliance to determine when AI tools should be integrated into clinical care [50]. Utah has taken a related route, producing a regulatory review that informed legislation and best-practice guidance. Its lessons include stakeholder divergence, evolving risk-benefit considerations, and the need for adaptive regulation, with a preference for evidence-based protocols, continuous monitoring, and holistic, inclusive policymaking [51]. A recent policy brief on generative AI as a de facto mental health provider outlines both opportunities and risks and provides actionable recommendations for users, clinicians, developers, and regulators, arguing for regulation proportional to the fact that these platforms now serve as mental health resources for millions [2]. Internationally, ethical governance is still uneven. Work on global governance challenges emphasizes that human vulnerability, especially in mental health contexts, requires ethical risk analysis and global coordination that current instruments do not fully provide [52]. Systematic reviews of methodologies, evaluation frameworks, and ethical safeguards for LLM chatbots and agentic AI in mental health counseling have concluded that current evidence is limited by heterogeneous study designs, minimal external validation, and inconsistent reporting of safety and governance practices, with no included study reporting registered randomized controlled trials or independent clinical validation in real-world care settings [12]. Future work is urged to prioritize clinically grounded evaluation frameworks, transparent reporting of model and prompt configurations, and stronger validation using standardized outcomes to support safe, reliable, and regulatory-ready deployment [12]. Taken together, the regulatory picture is one of catch-up. Some jurisdictions are drawing bright lines; others are relying on general-purpose consumer protection law that was never designed for autonomous digital counselors. In this environment, the practical burden of #safety and ethics remains disproportionately on developers, deploying institutions, and users themselves. 10. Vulnerable Populations and Global Equity The promise of agentic AI counselors is loudest where care is scarcest. Digital psychological interventions have shown modest but genuine benefits for mental health problems in low- and middle-income countries, though methodological quality varies and long-term follow-up is limited [53, 54]. AI chatbots are often positioned as tools that can extend care to populations that have never had access to specialist services. The concern is that the same populations most in need are also least well positioned to correct the system. Justice-oriented analyses of vulnerable populations in AI-mediated psychotherapy have argued that developers must design specifically for the needs of these users rather than assume that mainstream training data will generalize [37]. Practitioner-informed frameworks describe the disparity plainly: users who are knowledgeable enough to correct LLM outputs are at an advantage, while others are more likely to suffer from clinically inappropriate responses [26]. The problem is compounded when systems are deployed cross-culturally. Reviews document that cultural context strongly shapes both privacy expectations and what users experience as helpful therapeutic dialogue [44]. Ethical frameworks proposed for human-bot psycho-counseling explicitly reject the idea of a single universal code and argue that ethical principles must be tailored to context [38]. Genetic counseling scholarship has raised similar concerns about generative AI's fit for consequential clinical conversations that require sensitivity to family, culture, and lived experience [55]. Adolescents present a particular case. AI as a mental health therapist for adolescents has been analyzed both for its potential and for concerns about developmental appropriateness, exposure to unfiltered content, and the challenge of consent [56]. Meta-analytic evidence in the 15-39 age range indicates measurable benefits but underscores that engagement, repetitive content, and technical issues remain barriers [15]. Older adults are another underexplored group. A 2025 meta-analysis of digital mental health interventions for older adults synthesized randomized controlled trials and reported meaningful improvements, alongside caveats about digital literacy and engagement [57]. Postpartum populations similarly show modest but measurable benefits from digital interventions [58]. An equity-based reading of this evidence produces a mixed conclusion. Agentic AI counselors could genuinely extend care to underserved populations, but only if their design, evaluation, and governance treat these populations as the primary users rather than an afterthought. Without such deliberate design, the technology risks reproducing or worsening existing #health_inequities. 11. Recommendations for Responsible Development Several complementary recommendations emerge from the recent literature. Center clinical science. A responsible development roadmap for clinical LLMs in psychotherapy has argued that development must center clinical science, involve robust interdisciplinary collaboration, and attend to issues such as assessment, risk detection, transparency, and bias [3]. In particular, the field should draw parallels with the development of autonomous vehicle technology, in which progressive stages of automation each require independent validation and fail-safe architectures [3]. Adopt fiduciary design. As argued above, fiduciary design offers a unified paradigm for #AI_trust and accountability, extending professional norms of loyalty, care, and candor into system design [31]. This includes constraints on optimization objectives, mandatory disclosure of the system's nature, and structural limits on secondary use of user data. Layer safety mechanisms. Systematic reviews of safety mechanisms and risk mitigation in generative AI mental health chatbots recommend a sociotechnical approach that integrates technical safeguards with #user_co_design, procedural controls, and human oversight [39]. Layered safety architectures combining content filters, risk classifiers, and rule-based algorithms outperform any single mechanism [39]. Crisis referral protocols must be evaluated, not merely declared. Design for #therapeutic_alliance honestly. Because alliance predicts outcome even with autonomous agents [18], designers should build for it deliberately, but without misrepresenting the system's nature. Externalization techniques such as visual and dialogic externalization have been proposed as ways to use generative AI as an "artificial third" that enhances rather than supplants the human therapist relationship, with structured guidelines for responsible integration [59]. Regulate proportionally. Regulatory frameworks should be evidence-based, adaptive, and proportional to the risks these systems carry [51, 49]. Illinois's Public Act 104-0054 and Utah's regulatory review provide early domestic templates, and cross-jurisdictional coordination is needed to prevent forum shopping and race-to-the-bottom deployment [50, 51]. Protect mental privacy at architecture level. Federated learning and privacy-preserving model architectures are practical routes to reduce the surface area for privacy harm [47, 48]. Recognition of #mental_privacy and mental integrity as distinct rights would provide firmer legal ground [45, 46]. Standardize evaluation. Reviews of LLM based counseling chatbots have found low external validation and inconsistent reporting on ethics and safety [12]. Standardized outcome measures, prespecified safety endpoints, and mandatory reporting of adverse events would allow honest comparison across products. Preserve human clinical roles. Frameworks for the integration of conversational AI in psychotherapy propose several models of #clinician_AI collaboration and argue that if the technology is not safe it should not be used, and if it is not trusted, it will not be [7]. Human oversight is not a courtesy; it is a design requirement. Match the risk profile. For low-severity, subclinical presentations, evidence supports autonomous systems as a useful entry point into care [16, 15]. For moderate to severe presentations, current evidence supports adjunctive use only, with human clinicians retaining diagnostic and safety authority. 12. Limitations of the Current Evidence Several limitations of the current literature must be acknowledged for students working with these sources. First, publication bias is likely substantial. Trials of AI chatbots that did not show positive results are underrepresented, and the 2024 meta-analysis explicitly adjusted for publication bias, with resulting effect sizes smaller than raw estimates [14, 16]. Second, most trials have short durations and lack long-term follow-up, so durability claims are largely unsupported [14]. Third, most participants are drawn from student, volunteer, or online recruitment pools, and generalization to severe or comorbid clinical presentations is limited [16, 18]. Fourth, control conditions vary widely. Some trials use waitlist controls that inflate apparent effect sizes; others compare against general-purpose chatbots or #internet_CBT that themselves have positive effects [19]. Fifth, several important claims come from scoping reviews and commentary rather than randomized evidence, and safety data remain uneven across products [39, 33]. Sixth, real-world implementation trials suggest that laboratory efficacy does not automatically translate to routine care [24]. These limitations do not invalidate the field. They shape how confident a careful reader should be in specific claims. The strongest supported claim is that #agentic and generative AI counselors can produce small-to-moderate short-term reductions in depression and anxiety in mildly clinical populations. Everything larger than that requires more caution. 13. Conclusion An honest reading of the recent literature supports a middle position. #Agentic_AI counselors are not the transformative replacement for human care that some early promotional writing suggested, nor are they the unqualified danger that some critics have feared. They are, at their best, useful adjuncts that can reduce short-term symptoms of depression and anxiety in mildly to moderately symptomatic users, extend access to underserved populations, and support engagement in evidence-based techniques. They are, at their worst, systems that simulate empathy without possessing it, that store the most intimate data users produce, that fail in predictable ways in crisis, and that operate outside the professional accountability structures the mental health field has built over a century. The three questions with which this article began have partial answers. On efficacy, the evidence supports small-to-moderate short-term effects on depression and anxiety, mediated in part by #digital_therapeutic_alliance, with limited evidence on durability, severity generalization, and comparative effectiveness against human care. On fiduciary duty, a growing scholarly consensus argues that systems occupying positions of trust in the mental health domain should be held to standards analogous to those imposed on human professionals, translated into design constraints, accountability structures, and disclosure obligations. On ethical boundaries, the most consistently cited limits concern safety in crisis, protection of mental privacy, avoidance of deceptive empathy and sycophancy, protection of user autonomy, and equitable treatment of vulnerable populations. The task facing the next generation of researchers, clinicians, and policymakers is not to accept or reject #digital_co_counselors as a category. It is to specify, empirically and normatively, the conditions under which they help, the conditions under which they harm, and the accountability structures required to keep the two apart. That work is well underway, but it is far from finished, and the pace of technological change is unlikely to slow in the interim. For students entering this field, the practical guidance is straightforward. Read the primary trials, not the press releases. Track effect sizes, follow-up durations, and control conditions rather than headline claims. Pay attention to who was excluded from the sample, because that is often where the ethical action lies. Insist on transparent safety data and on explicit fiduciary structures before endorsing any deployment at scale. And remember, in every design decision, that the users of these systems are usually people in real distress, whose #trust and #vulnerability deserve the same level of professional seriousness a human clinician would offer. Hashtags #agentic_AI #digital_mental_health #co_counselor #chatbot_therapy #LLM_ethics #fiduciary_duty #therapeutic_alliance #mental_privacy #AI_safety #generative_AI #psychotherapy_AI #AI_regulation #conversational_AI #responsible_AI #mental_health_innovation References Kaur, J., Nasir, M. J., Velez, L., Jain, L., and Raza, M. (2025). Too good to be true? Exploring the role of artificial intelligence chatbots in treating depression and anxiety. Discover Mental Health. Refoua, E., Gigi, K., Levkovich, I., Hadar Shoval, D., Elyoseph, Z., Tsafrir, I., Pen, O., Angert, T., and Haber, Y. (2026). Generative AI as a de facto mental health provider: a policy brief and urgent call for regulation. Frontiers in Digital Health. Stade, E. C., Stirman, S., Ungar, L., Boland, C., Schwartz, H. A., Yaden, D., Sedoc, J., DeRubeis, R. J., Willer, R., and Eichstaedt, J. (2024). Large language models could change the future of behavioral healthcare: a proposal for responsible development and evaluation. npj Mental Health Research, 3, 12. Fitzpatrick, K. K., Darcy, A., and Vierhile, M. (2017). Delivering Cognitive Behavior Therapy to Young Adults With Symptoms of Depression and Anxiety Using a Fully Automated Conversational Agent (Woebot): A Randomized Controlled Trial. JMIR Mental Health, 4(2), e19. Inkster, B., Sarda, S., and Subramanian, V. (2018). An Empathy-Driven, Conversational Artificial Intelligence Agent (Wysa) for Digital Mental Well-Being: Real-World Data Evaluation Mixed-Methods Study. JMIR mHealth and uHealth, 6(11), e12106. Gaffney, H., Mansell, W., and Tai, S. (2019). Conversational Agents in the Treatment of Mental Health Problems: Mixed-Method Systematic Review. JMIR Mental Health, 6(10), e14166. Miner, A. S., Shah, N., Bullock, K., Arnow, B., Bailenson, J., and Hancock, J. (2019). Key Considerations for Incorporating Conversational AI in Psychotherapy. Frontiers in Psychiatry, 10, 746. Ravikumar, V. (2024). Agentic AI Systems for Personalized Mental Health Interventions. International Journal of Emerging Technologies and Innovative Research. Pattar, V., et al. (2025). Mindcare: A Multi-Agent AI Architecture for Personalized and Responsible Mental Health Support. Proceedings of the International Conference on Artificial Intelligence in Healthcare. Ma, A., et al. (2026). From Tool to Agent: A Semi-Systematic Review of Human-AI Alignment and a Proposed Tiered Healing Ecosystem for Mental Health. Journal of Medical Internet Research. Nur, J., Tan, K., and Wiputra, R. (2025). AI Agents in Mental Health Treatment: A Systematic Review. Journal of Digital Health Research. Cho, H., Zheng, K., Wang, J., and Hu, D. (2025). Large Language Model-Based Chatbots and Agentic AI for Mental Health Counseling: Systematic Review of Methodologies, Evaluation Frameworks, and Ethical Safeguards. JMIR Mental Health. Olisaeloka, L., Richardson, C. G., Wang, A. Y., Munthali, R. J., and Vigo, D. V. (2026). Safety Mechanisms and Risk Mitigation in Generative AI Mental Health Chatbots: A Systematic Scoping Review. npj Digital Medicine. Zhong, W., Luo, J., and Zhang, H. (2024). The therapeutic effectiveness of artificial intelligence-based chatbots in alleviation of depressive and anxiety symptoms in short-course treatments: A systematic review and meta-analysis. Journal of Affective Disorders, 356, 459-469. DOI: 10.1016/j.jad.2024.04.057. Feng, X., Tian, L., Ho, G. W., Yorke, J., and Hui, V. (2025). The Effectiveness of AI Chatbots in Alleviating Mental Distress and Promoting Health Behaviors Among Adolescents and Young Adults: Systematic Review and Meta-Analysis. Journal of Medical Internet Research. Feng, Y., Hang, Y., Wu, W., Song, X., Xiao, X., Dong, F., and Qiao, Z. (2024). Effectiveness of AI-Driven Conversational Agents in Improving Mental Health Among Young People: Systematic Review and Meta-Analysis. Journal of Medical Internet Research, 26, e60432. Heinz, M. V., Mackin, D. M., Trudeau, B. M., Bhattacharya, S., Wang, Y., Banta, H. A., Jewett, A. D., Salzhauer, A. J., Griffin, T. Z., and Jacobson, N. C. (2025). Randomized Trial of a Generative AI Chatbot for Mental Health Treatment. NEJM AI. Shoshani, A., Gurfinkel, B., Kor, A., Ben-Haim, Y., Kanarek, O., Segev, R., Shafir, O., and Arbel, R. (2026). Efficacy of a Conversational AI Agent for Psychiatric Symptoms and Digital Therapeutic Alliance. JAMA Network Open. Kuta, B., Novak, L., Zidkova, R., Furstova, J., Malinakova, K., de Winter, A. F., and Husek, V. (2025). Effectiveness of a Fully Automated Mobile Therapeutic Versus a General Chatbot in Reducing Depression and Anxiety and Improving Well-Being: Feasibility Randomized Controlled Trial. JMIR Formative Research. Campellone, T., Flom, M., Montgomery, R. M., Bullard, L., Pirner, M., Pavez, A., Morales, M., Harper, D., Oddy, C., O'Connor, T., Daniels, J., Eaneff, S., Forman-Hoffman, V., Sackett, C., and Darcy, A. (2025). Safety and User Experience of a Generative Artificial Intelligence Digital Mental Health Intervention: Exploratory Randomized Controlled Trial. JMIR Mental Health. Farzan, M., Ebrahimi, H., Pourali, M., and Sabeti, F. (2024). Artificial Intelligence-Powered Cognitive Behavioral Therapy Chatbots: A Systematic Review. Iranian Journal of Psychiatry. Nur, J., Tan, K., and Wiputra, R. (2025). AI Agents in Mental Health Treatment: A Systematic Review. Journal of Digital Health Research. Surana, K., Sharma, A., Tibrewal, V., and Kushwaha, R. (2025). GenAI-based Mental Health Chatbots: Efficacy, Personalization and Ethical Considerations. International Journal of Human-Computer Interaction. Meinert, E., Milne-Ives, M., Taylor, E., Inkster, B., Paik, A., Ananthakrishnan, A., Orr, M., Costelloe, C., and Shankar, R. (2026). Real-World Testing of an Artificial Intelligence Conversational Agent as an Early Intervention and Support Tool in the Mental Health Referral Care Pathway: A Mixed-Methods Study. JMIR Human Factors. Beatty, C., Malik, T., Meheli, S., and Sinha, C. (2022). Evaluating the Therapeutic Alliance With a Free-Text CBT Conversational Agent (Wysa): A Mixed-Methods Study. Frontiers in Digital Health, 4, 847991. Iftikhar, Z., Xiao, A., Ransom, S., Huang, J., and Suresh, H. (2025). How LLM Counselors Violate Ethical Standards in Mental Health Practice: A Practitioner-Informed Framework. Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society. Sedlakova, J., and Trachsel, M. (2022). Conversational Artificial Intelligence in Psychotherapy: A New Therapeutic Tool or Agent? The American Journal of Bioethics, 23(5), 4-13. On, Y., et al. (2024). Psychotherapy Based on the Large Language Models: On the Aspect of the Theory of Mind, a Narrative Review. Psychiatry Investigation. Wadden, J. J. (2023). The Postphenomenological Impact of Conversational Artificial Intelligence on Autonomy and Psychological Integrity. The American Journal of Bioethics, 23(5), 89-91. Olisaeloka, L., Nunez, J.-J., Vigo, D. V., and Ng, R. (2026). Artificial intelligence (AI) psychosis: mechanisms, clinical risks and safety considerations in generative AI chatbots. The Lancet Digital Health. Erickson, J. (2026). Who Does Your AI Work For? Designing Conversational Agents as Digital Fiduciaries. Yale Journal of Law and Technology. Martinez-Martin, N. (2021). Minding the AI: Ethical Challenges and Practice for AI Mental Health Care Tools. In: Jotterand, F. and Ienca, M. (eds), Artificial Intelligence in Brain and Mental Health: Philosophical, Ethical & Policy Issues, Springer. Rahsepar Meadi, M., Sillekens, T., Metselaar, S., van Balkom, A., Bernstein, J., and Batelaan, N. (2025). Exploring the Ethical Challenges of Conversational AI in Mental Health Care: Scoping Review. Journal of Medical Internet Research. Grodniewicz, J. P., and Hohol, M. (2023). Therapeutic Conversational Artificial Intelligence and the Acquisition of Self-understanding. The American Journal of Bioethics, 23(5), 59-61. Bertina, A., and Shakeri, S. (2026). Irresponsible Counselors: Large Language Models and the Loneliness of Modern Humans. AI & Society. Wang, Y. (2025). The Risks and Potential of Large Language Models in Mental Health Care. Undisturbed: Interdisciplinary Studies Journal. Knox, B., et al. (2023). Justice, Vulnerable Populations, and the Use of Conversational AI in Psychotherapy. The American Journal of Bioethics, 23(5), 87-89. Ma, L., Zhao, T., Qiu, H., and Lan, Z. (2024). No General Code of Ethics for All: Ethical Considerations in Human-bot Psycho-counseling. AI and Ethics. Olisaeloka, L., Richardson, C. G., Wang, A. Y., Munthali, R. J., and Vigo, D. V. (2026). Safety Mechanisms and Risk Mitigation in Generative AI Mental Health Chatbots: A Systematic Scoping Review. npj Digital Medicine. Rolvien, L., Kruse, L., Rings, S., Zimmer, C.-N., Schauenburg, G., Thams, F., Brahler, A., Rudschies, C., Schneider, I., Moritz, S., Steinicke, F., and Gallinat, J. (2026). Intelligent virtual agents in psychotherapy: a safety evaluation across high-risk mental health scenarios. BMC Psychiatry. MacIntyre, M. R. (2026). When the therapist hallucinates: AI, psychedelics, and the risks of unsupervised digital mental health care. Journal of Psychedelic Studies. Stiefel, S. (2018). The Chatbot Will See You Now: Mental Health Confidentiality Concerns in Software Therapy. Journal of Law, Technology & Policy. Ahmed, S., Zannat, R., Asaduzzaman, M., Faisal, H. M., Rashid, M., and Moni, S. F. (2025). Boundary Turbulence in Digital Spaces: Privacy Management in AI-Powered Mental Health Platforms. International Communication Research Reports. Ahmed, S., and Asaduzzaman, M. (2026). The Global Privacy Frontier: Bridging User Expectations and AI in Mental Health Technology. Journal of Communication Technology. Szoszkiewicz, L., and Yuste, R. (2025). Mental privacy: navigating risks, rights and regulation. EMBO Reports, 26. DOI: 10.1038/s44319-025-00505-6. Farinella, F., et al. (2024). Neurorights: Time to Discuss Rights to Mental Privacy and Integrity. European Journal of Law and Technology. Gangwar, A., et al. (2025). Federated Learning for Privacy Preserving AI in Mental Health Applications. IEEE Access. Mandal, A., et al. (2025). Towards privacy-aware mental health AI models. Proceedings of the ACL Workshop on NLP for Positive Impact. Joshi, S. (2025). Regulatory Frameworks for Generative AI Enabled Digital Mental Health Devices: Safety, Transparency, and Post-Market Oversight. Journal of Healthcare Informatics Research. Szoke, D. R., Pridgen, S. A., and Held, P. (2025). Artificial Intelligence in Mental Health Services Under Illinois Public Act 104-0054: Legal Boundaries and a Framework for Establishing Safe, Effective AI Tools. Journal of Clinical Psychology. de Lacy, N., and Boyd, Z. (2026). The doctor is not in, but the Chatbot is: Utah's experience regulating mental health AI. Health Affairs Scholar. Ansari, K. (2025). Human Vulnerability, Machine Law: Ethical Risks and Global Governance Challenges in Generative AI. Global Policy. Fu, Z., et al. (2020). Effectiveness of digital psychological interventions for mental health problems in low-income and middle-income countries: a systematic review and meta-analysis. The Lancet Psychiatry, 7(10), 851-864. Karyotaki, E., et al. (2023). Digital interventions for common mental disorders in low- and middle-income countries: A systematic review and meta-analysis. PLOS Digital Health. Meekins-Doherty, L., et al. (2025). Generative AI and the profession of genetic counseling. Journal of Genetic Counseling. Opel, D. J., et al. (2023). AI as a Mental Health Therapist for Adolescents. JAMA Pediatrics, 177(12), 1253-1254. Wang, Y., et al. (2025). Digital Mental Health Interventions for Older Adults: A Meta-Analysis of Randomized Controlled Trials. Journal of the American Geriatrics Society. Lewkowitz, A., et al. (2023). The effect of digital health interventions for postpartum depression or anxiety: a systematic review and meta-analysis of randomized controlled trials. American Journal of Obstetrics and Gynecology. Haber, Y., Shoval, D., Levkovich, I., Yinon, D., Gigi, K., Pen, O., Angert, T., and Elyoseph, Z. (2025). The externalization of internal experiences in psychotherapy through generative artificial intelligence: a theoretical, clinical, and ethical analysis. Frontiers in Digital Health.

The global expansion of professional #counselling has produced a striking pattern. Training programs in Nairobi, Manila, Toronto, and Riyadh increasingly look alike. They use similar textbooks, teach similar competencies, follow similar practicum hours, and measure student learning against similar outcome rubrics. This paper argues that this convergence is not accidental. It reflects a form of institutional isomorphism, a process in which organizations that operate in the same field come to resemble one another because of shared pressures rather than shared cultures. Drawing on the classical framework of coercive, mimetic, and normative isomorphism, and updating it with recent scholarship on transnational governance in the helping professions, the paper examines how international #accreditation bodies, cross border quality assurance metrics, and portable licensure agreements are quietly reshaping clinical therapeutic practice. The analysis shows that while these mechanisms improve consistency, portability, and public safety, they also flatten local knowledge, marginalize indigenous healing traditions, and privilege a specific, largely Anglo American, model of the counselling relationship. The paper closes by suggesting how future accreditation systems might preserve professional rigor without demanding cultural uniformity, and it proposes a research agenda for students who wish to study the politics of standard setting in the helping professions. Keywords: #institutional_isomorphism, #global_counselling, #accreditation, #quality_assurance, #cultural_competence, #clinical_practice, #professional_regulation, #cross_border_licensure 1. Introduction Walk into a #counsellor training program in almost any country in 2026 and you will find a familiar scene. Students discuss the therapeutic alliance. They practice reflective listening in role plays. They study cognitive behavioural techniques, person centred principles, and a brief unit on cultural humility. Their syllabus lists learning outcomes, mapped to a competency framework, mapped to an #accreditation standard, mapped to a national qualifications framework, mapped in turn to a set of international reference points. A student in Manila and a student in Manchester may sit thousands of kilometres apart, but their weekly reading lists overlap more than one might expect. This is not simply the natural spread of good ideas. It is the visible surface of a much deeper process. The #helping_professions have entered a period in which shared standards, portable credentials, and global rubrics increasingly define what counts as competent clinical work. National regulators borrow the language of international bodies. Universities align their curricula with cross border accreditation manuals. Insurers and public health systems demand that therapists prove their skills against standardized #competency frameworks. Even in countries that never signed a formal treaty on mental health workforce, the pull toward a common professional template is strong. Sociologists have a name for this. They call it #institutional_isomorphism. The concept comes from a classic paper by DiMaggio and Powell, who noticed that organizations in the same field, whether hospitals, schools, or museums, tend to become similar over time even when the pressures on them look different (DiMaggio and Powell, 1983). Recent scholarship has extended their framework into the transnational domain, showing that professional fields now converge across borders because of shared regulators, shared training pipelines, and shared anxieties about legitimacy (Djelic and Sahlin Andersson, 2021; Meyer, 2022). This paper applies that lens to the field of counselling and psychotherapy. It asks a simple question. Why do therapeutic practices in very different societies increasingly look the same, and what are the costs and benefits of that convergence for clinical work, for students, and for the communities that counsellors are meant to serve. The argument is that #international_accreditation and #quality_assurance metrics are the main engines of #homogenization in the field. They do this in three ways, which correspond to the three classical types of isomorphism. They coerce through legal and financial mandates. They invite mimicry through the language of best practice. They normalize through shared training pipelines. Each of these mechanisms is powerful. Together they produce a global counselling culture that is remarkably consistent, sometimes to the detriment of clinical wisdom. The paper is written for students who are entering the counselling profession at a moment when its rules are being rewritten. It aims to give them conceptual tools to see the invisible architecture around their training, and to raise questions they can carry into their thesis work, their supervision, and eventually their clinical careers. It is not a call to reject #standards. Standards protect clients. It is a call to notice how standards travel, whose interests they encode, and where they should be resisted in the name of good practice. The paper is organized as follows. Section two reviews the theoretical foundations of institutional isomorphism and how recent scholarship has extended the framework to cross border professional fields. Section three describes the conceptual method used in the paper and the boundary of the analysis. Sections four through six examine coercive, mimetic, and normative isomorphism in turn, with examples drawn from #counsellor_education, licensure portability, and outcome measurement. Section seven considers what is lost as counselling homogenizes, especially with respect to indigenous knowledge and low resource settings. Section eight looks at counter currents and points of resistance, including recent work on decolonising #therapy. Section nine sets out implications for training and supervision. Section ten concludes with a research agenda. 2. Theoretical Framework 2.1 The classical model The starting point for this paper is the framework developed by DiMaggio and Powell in the early 1980s, which remains the most influential account of organizational similarity. They argued that once a field becomes structured, meaning that its organizations share a common purpose, exchange information, and recognize one another as peers, powerful forces begin to push members toward looking alike. Three mechanisms drive this pull. Coercive isomorphism arises when organizations must comply with legal, regulatory, or financial rules imposed by more powerful actors, whether states, funders, or professional bodies. Mimetic isomorphism arises under uncertainty, when organizations copy peers they see as successful in order to reduce risk. Normative isomorphism arises through professionalisation, when members of an occupation are trained in the same institutions, absorb the same conceptual vocabulary, and travel through the same networks (DiMaggio and Powell, 1983). Although the original framework was written in the language of American organizational sociology, later scholars extended it to make sense of transnational phenomena. Djelic and Sahlin Andersson showed that #standards themselves have become a global governance form, quietly shaping fields as diverse as accounting, environmental regulation, and higher education (Djelic and Sahlin Andersson, 2021). Meyer and colleagues extended the world society tradition, arguing that even professions with strong local roots absorb globally circulating scripts about what a modern practitioner should know and how they should work (Meyer, 2022; Bromley and Meyer, 2021). 2.2 Extensions to the helping professions Applying this literature to counselling and psychotherapy is fairly new. For much of the twentieth century, mental health care was understood as a national or even local matter, closely tied to specific philosophical schools and religious traditions. That changed after the 1990s. The global mental health movement, the expansion of evidence based #psychotherapy, and the rise of transnational professional associations all pushed the field toward a more unified identity (Watters, 2021; Mills and Fernando, 2022). Recent studies show that counselling and #psychotherapy training programs across the world are increasingly aligned with a small number of accreditation templates. Fickling and colleagues found strong convergence in the way counsellor competencies are defined across accredited programs in North America, Europe, and parts of Asia (Fickling, Tangen, Graden, and Grays, 2021). Kaplan and colleagues traced how the international portability project in counselling has pushed national bodies to adopt shared definitions of scope of practice (Kaplan, Francis, Hermann, Baca, and Goodnough, 2021). Watkins and colleagues documented the same pattern in psychotherapy #supervision, where a global model of competency based supervision has become the default (Watkins, 2022). At the same time, scholars working from postcolonial and critical positions have raised questions about what the convergence excludes. Mills and Fernando argued that the global mental health apparatus tends to export a particular clinical worldview, one that assumes an autonomous individual, a bounded self, and a talking cure, into contexts where those assumptions do not fit (Mills and Fernando, 2022). Kirmayer and Jarvis pointed to a growing tension between #evidence_based practice as defined in high income settings and the plural healing traditions of much of the world (Kirmayer and Jarvis, 2022). 2.3 The framework used in this paper The paper synthesizes these strands into a working framework. Institutional isomorphism in global counselling is treated as a multilevel process in which: First, transnational bodies, whether formal like the International Association for Counselling or informal like large research networks, articulate templates for what counselling should be. These templates travel through publications, conferences, and training exchanges. Second, national regulators and universities, under pressure to demonstrate quality and to make their graduates internationally mobile, translate the templates into local rules. This is where the mechanisms of coercive, mimetic, and normative isomorphism operate most visibly. Third, individual practitioners internalize the templates through their training, their supervision, and their professional identity. By the time they are in the room with a client, the templates feel less like external rules and more like common sense. The task of the paper is to make this process visible. 2.4 Why isomorphism matters for a young profession One further piece of theory needs to be added before moving to the analysis. Counselling is a relatively young profession compared with medicine or law. Its boundaries are contested. Its knowledge base is broad rather than deep. Its practitioners come from many disciplinary backgrounds, including psychology, social work, education, pastoral care, and public health. In fields that are still consolidating, the pressure to appear legitimate is especially strong. Legitimacy comes from looking like other legitimate actors. This is why young professions often adopt the trappings of older ones, from accreditation bodies to codes of ethics to specialised journals, sometimes before the underlying knowledge base is fully settled (Bromley and Meyer, 2021; Djelic and Sahlin Andersson, 2021). This dynamic helps to explain why counselling has been unusually receptive to the machinery of #standards. In a field where the outcomes of therapy are hard to measure and where multiple schools of thought compete for authority, an external standard offers something valuable, namely a public marker of competence. The value is real. But the same standards also freeze the field around a particular version of itself. Understanding this trade off is the beginning of a mature relationship with the standards regime. 3. Method This is a conceptual paper. It does not present new empirical data. Instead, it synthesizes recent literature on #accreditation, quality assurance, and professional regulation in counselling with the theoretical apparatus of institutional isomorphism. The paper draws on peer reviewed articles and books published mostly between 2021 and 2025, supplemented by a small number of older classical works that are still cited in the field. Sources were identified through the standard academic databases, with search terms including counsellor accreditation, counsellor education standards, cross border licensure, competency based supervision, and cultural competence in psychotherapy. Grey literature such as unpublished policy documents is not cited, in keeping with the goal of building on stable scholarly foundations. The boundary of the paper is important. The analysis focuses on formal counselling and psychotherapy training as delivered through accredited university programs and post qualification professional bodies. It does not examine informal helping, peer support, or lay counselling systems in detail, even though those systems serve a large share of the world population. The paper also focuses on English language literature, which reflects both a limitation and, importantly, a feature of the very homogenization it examines. 4. Coercive Isomorphism in Global Counselling 4.1 Regulation and legal recognition Coercive isomorphism is the most direct way in which counselling practices become similar across borders. It occurs when governments, funders, or professional bodies require organizations to comply with specific rules as a condition of operating. In counselling, coercion appears in three main forms. The first is state #regulation of the counsellor title. In many jurisdictions the title of counsellor or psychotherapist is protected by law, and only those who hold a recognized qualification can practice under it. When national regulators design these laws, they rarely start from scratch. They look to peer jurisdictions and international bodies for guidance on what a counsellor should know, how they should be trained, and how their scope should be defined (Kaplan et al., 2021). The second is #accreditation of training programs. Universities that wish to offer a counselling degree in a regulated market must satisfy an accreditation body that their curriculum, faculty, and clinical placements meet a set of standards. These bodies, whether national or international, publish detailed manuals that specify course content, contact hours, supervisory ratios, and outcome metrics. Non compliance means the loss of recognition, which in turn means graduates cannot register or find work (Fickling et al., 2021). The third is #payer regulation. Insurers, public health systems, and non governmental organizations that pay for counselling services often require providers to meet specific standards, use specified assessment tools, and deliver a specified number of sessions. Because these payers control the flow of clients, their rules act as a form of coercion even when no law directly requires compliance (Fairburn and Patel, 2022). 4.2 Case example, credentialing across borders Consider the example of a counsellor trained in one country who wishes to practice in another. In the past this required lengthy examination and often extra coursework. Under the growing pressure for a mobile mental health workforce, professional bodies have signed portability agreements that spell out equivalent standards. To be eligible under such agreements, a training program in the sending country must show that it meets the receiving country standard. In practice, this means that many countries have quietly rewritten their standards to match the requirements of the largest and most influential #accreditation systems (Kaplan et al., 2021). This is coercion in a soft form. No one forces a Kenyan or Filipino training program to align with a North American manual. But if a program wants its graduates to be eligible for licensure in the wealthier markets where they might migrate, alignment becomes a practical necessity. Over time, the pull of these markets shapes the design of programs even in countries whose own citizens have very different mental health needs. 4.3 Consequences The coercive channel has some clear benefits. It provides a floor of #clinical_competence that protects clients from unqualified practitioners. It makes it easier to move between jurisdictions in a globalized labour market. It gives universities a stable framework on which to build programs. But it also has costs. When accreditation manuals define counselling in narrow terms, alternative approaches struggle to gain footing. When outcome metrics reward short structured therapies, longer relational work becomes harder to fund. When the standards implicitly assume a certain kind of client, one who speaks the language of goals and outcomes, counsellors trained under those standards may struggle when the client in front of them does not fit that mould. These consequences are not the fault of any single body. They are the systemic effect of a coercive push toward a common template. 4.4 The role of insurance and public financing One aspect of coercive isomorphism deserves closer attention because it operates at a distance from formal regulation. Insurers and public health systems increasingly shape counselling practice through what they pay for. Reimbursement rules typically require a diagnosis using an internationally recognized #classification system, a treatment plan tied to that diagnosis, and outcome tracking through a small number of validated questionnaires. In systems where most therapy is financed through insurance or a national health service, these rules do more to standardize practice than any regulator (Fairburn and Patel, 2022; Barkham et al., 2021). The effect is felt in small daily choices. A counsellor might prefer to work with a client for a longer period, without a formal diagnosis, using a relational approach. If the payer will only fund short structured therapy for a diagnosable condition, the counsellor either adapts or loses the case. Over time, thousands of such small choices produce a workforce whose practice patterns look similar even when their theoretical orientations differ. This form of coercion is not usually described as isomorphism in policy documents. It is presented as accountability, or value for money, or evidence based commissioning. But its effect on the shape of practice is unmistakable. Students who wish to understand the mechanics of #homogenization in counselling should look carefully at how their local financing system rewards some kinds of work and quietly discourages others. 5. Mimetic Isomorphism in Global Counselling 5.1 Uncertainty and the temptation to copy If coercion pushes counsellors toward similarity through rules, mimicry pulls them toward similarity through example. DiMaggio and Powell argued that organizations facing uncertainty about how to succeed often copy peers they perceive as successful. Counselling is a field with genuine uncertainty. Clients present with a wide range of problems. Interventions do not always work in expected ways. Programs must prove themselves to regulators, funders, and prospective students. Under such conditions, imitation is attractive. In counselling the most common form of mimicry is the adoption of curricular models and clinical protocols from a small number of high visibility programs and manuals. A new master level counselling program in one country might, when designing its curriculum, look at the curriculum of a leading program in another country. It might adopt the same textbook, structure its practicum in the same way, and even hire graduates of that program to teach (Neukrug, 2022; Corey, 2023). 5.2 Manualised therapies and their spread A related form of mimicry is the global spread of manualised therapies. #Cognitive_behavioural_therapy, dialectical behaviour therapy, motivational interviewing, and eye movement desensitization and reprocessing are the best known examples. Each has a manual that spells out what the therapist should do session by session. Each has a certification pathway that produces a growing supply of trained practitioners across many countries (Hofmann and Hayes, 2022). The appeal of manualised therapy is understandable. It gives new counsellors a clear script. It provides a shared vocabulary for #supervision. It generates the kind of outcome data that impresses regulators and funders. Under uncertainty, manualised therapy offers a form of collective certainty (Barkham, Lutz, and Castonguay, 2021). But mimicry is not the same as understanding. A protocol developed in an urban western clinic may not translate cleanly into a rural setting where clients do not have the private space to complete homework assignments, or where the emotional vocabulary of the manual does not fit the local language. When counsellors copy protocols without deeply adapting them, the manual becomes a ritual rather than a tool (Kirmayer and Jarvis, 2022; Watters, 2021). 5.3 Rankings, prestige, and the manufacture of best practice A subtle but powerful driver of mimicry is the language of best practice. When journals, textbooks, and professional bodies label a particular approach as evidence based or as best practice, they signal that copying it is a safe and prestigious choice. Programs and clinicians who deviate risk being seen as behind the times. Recent scholarship has drawn attention to how this label is constructed. Not every therapy that has a strong evidence base is equally promoted, and not every widely promoted therapy has an evidence base that generalizes across settings. The politics of what counts as #evidence_based practice is an active area of research, and it matters directly for the mimetic pressure on training programs (Fairburn and Patel, 2022; Barkham et al., 2021). 5.4 The circulation of clinical vocabulary Mimicry also operates at the level of vocabulary. A relatively small number of clinical terms, such as trauma, resilience, boundaries, self care, attachment, and mindfulness, have become the lingua franca of the global counselling field. These terms travel through popular books, social media, and continuing education workshops before they reach formal training. Students arrive at their first counselling class already fluent in them. This creates a shared conceptual space that supports international communication but that also crowds out alternative vocabularies drawn from local languages and traditions. The consequence is subtle. When a client describes their trouble in local terms, the counsellor may translate silently into the global vocabulary before formulating a response. Something is gained in that translation, namely a link to a wider clinical community. Something is also lost, namely the specific texture of the client's own way of naming their suffering (Kirmayer and Jarvis, 2022; Mills and Fernando, 2022). 5.5 Conferences, workshops, and the market for continuing education Another site of mimicry is the market for continuing education. Once qualified, counsellors are expected to accumulate professional development hours. A large industry of workshops, certifications, and online courses has grown up to meet this demand. The most successful providers operate across many countries and offer the same product in each. Their branding, marketing, and pedagogy set trends that shape what practitioners see as advanced practice. Because continuing education is a condition of licence renewal in many jurisdictions, this market is not peripheral. It is central to how the profession updates itself. And because the market rewards products that scale, it tends to reward standardization over local specificity (Barkham et al., 2021; Hofmann and Hayes, 2022). 6. Normative Isomorphism in Global Counselling 6.1 Professionalisation and shared training pipelines The third mechanism, normative isomorphism, is arguably the most enduring. It operates through the shared training and socialization of practitioners. When counsellors around the world learn from the same textbooks, attend the same conferences, and internalize the same professional identity, the field converges from the inside rather than the outside. The rise of a global counselling literature has been dramatic. Textbooks such as those by Corey, by Neukrug, and by Sommers Flanagan and Sommers Flanagan are used in classrooms across many countries (Corey, 2023; Neukrug, 2022; Sommers Flanagan and Sommers Flanagan, 2022). These books do more than teach techniques. They teach a way of thinking about the person, the problem, and the therapeutic relationship. Students who grow up reading them absorb assumptions about autonomy, boundaries, self disclosure, and the appropriate emotional distance between counsellor and client. Those assumptions then become the taken for granted background of their clinical work (Hodges, 2020; Corey, 2023). 6.2 Supervision and the making of a professional identity #Supervision plays a special role in normative isomorphism. It is the practice through which a novice counsellor is transformed into a member of the profession. Recent decades have seen the emergence of a global model of competency based supervision, in which supervisors assess trainees against defined competencies and use a common set of tools (Watkins, 2022; Bernard and Goodyear, 2021). This model has real strengths. It gives supervisors a language for feedback. It provides an accountability trail. It signals that supervision is a specialised skill in its own right, not merely a byproduct of clinical seniority. But when the same competencies are applied across cultures without adaptation, they can distort supervision. Feedback about eye contact or emotional expressiveness may reflect the norms of one culture rather than a universal standard of good practice. Supervisors trained in the dominant model may inadvertently penalize supervisees whose therapeutic style draws on other traditions (Watkins, 2022; Kirmayer and Jarvis, 2022). 6.3 Professional associations and mobility Professional associations amplify normative isomorphism. They convene conferences, publish journals, and offer #certification programs. Through their activities they define the boundaries of the profession and reward those who work within them. As associations increasingly operate across national lines, whether formally through international federations or informally through joint conferences, the templates they promote gain global reach (Kaplan et al., 2021; Djelic and Sahlin Andersson, 2021). For students the effect is subtle but powerful. Their sense of what a counsellor is, how a counsellor should speak, how a counsellor should dress, and even how a counsellor should decorate a therapy room is shaped by the images and norms that circulate through these networks. The result is a professional identity that is remarkably consistent across countries, even where the underlying cultures diverge sharply. 6.4 Ethics codes and the shared conscience of the profession Professional ethics codes deserve a note of their own. Across countries, ethics codes have converged around a common vocabulary of autonomy, beneficence, non maleficence, and justice, together with more specific commitments to informed consent, confidentiality, competence, and multiple relationships. A counsellor moving from one national association to another finds many of the same principles, sometimes phrased in identical language (Corey, 2023; Sommers Flanagan and Sommers Flanagan, 2022). This convergence has clear benefits for clients. It signals that certain protections are not optional. It gives practitioners a common reference when they face difficult decisions. But it also encodes a specific tradition, largely rooted in Western bioethics, into a global profession. Alternative ethical frames, such as those grounded in relational or communal traditions, sit awkwardly with the individualist emphasis of the dominant codes. Students who work across settings will benefit from studying not only the codes they must follow but also the ethical traditions those codes leave unspoken (Kirmayer and Jarvis, 2022). 6.5 Research training and the epistemic default A final normative channel is research training. Most counselling programs require students to study research methods and, in many cases, to complete a small research project. The methods that are taught are heavily weighted toward a narrow slice of the methodological universe, namely quantitative outcome studies, randomised trials, and psychometric measurement. Qualitative, participatory, and community based methods are usually treated as supplementary rather than central. The consequence is that graduates enter practice with a default view of what counts as knowledge. If the trials say so, it is real. If the community says so, it is interesting but softer. This epistemic default supports the wider isomorphism, because the therapies with the strongest trial evidence gain a privileged position in accreditation and reimbursement. Broadening the methodological training of counsellors is therefore not only a scholarly matter. It is a way to open the field to a wider range of knowledge (Barkham et al., 2021; Patel et al., 2022). 7. What is Lost, Homogenization and Local Knowledge 7.1 Indigenous healing traditions One of the most important costs of institutional isomorphism in counselling is the marginalization of indigenous and non western healing traditions. Every society has developed ways of responding to psychological suffering. These may involve religious ritual, community based practices, family led interventions, or storytelling traditions. In many places these approaches remain the first line of help for people in distress (Mills and Fernando, 2022; Kirmayer and Jarvis, 2022). When #accreditation standards define counselling in narrow terms, however, they implicitly say that these traditions are outside the professional field. Practitioners who blend counselling with indigenous methods may find that their work does not fit into the categories that regulators recognize. Graduates trained under global standards may return to their communities and find that the tools they learned do not translate into the emotional idiom of home. This is not a call to romanticize indigenous practice. Some indigenous methods are harmful, and clients deserve protection. It is a call to recognize that homogenization is not neutral. It privileges one clinical vocabulary while making others invisible. 7.2 The problem of #cultural_competence as a checklist Many accreditation systems now include a requirement that trainees be culturally competent. On the surface this looks like a response to the concerns above. In practice, however, cultural competence often becomes a checklist. Trainees learn a short list of facts about specific groups, complete a required course, and tick the box. The deeper questions of whose knowledge counts and how power moves through the therapy room are left aside (Kirmayer and Jarvis, 2022; Sue, Sue, Neville, and Smith, 2022). Recent scholarship has proposed the alternative concept of #cultural_humility, which emphasizes lifelong learning, self reflection, and attention to the power dynamics of the therapeutic relationship (Hook, Watkins, Davis, Owen, Van Tongeren, and Ramos, 2022). Cultural humility resists the checklist form because it cannot be reduced to a single course or a single assessment item. Its uptake into accreditation manuals has been uneven, in part because it is harder to measure than cultural competence. 7.3 Low resource settings and workforce shortages Institutional isomorphism has particular consequences for low resource settings. Countries with severe shortages of mental health workers rely on task sharing, in which lay providers deliver structured interventions under supervision. This approach has strong evidence for common conditions such as depression and anxiety (Patel, Saxena, and Lund, 2022; Fairburn and Patel, 2022). The rise of task sharing sits uneasily with the credentialing regime that shapes the formal counselling profession. If accreditation systems only recognize university trained #counsellors, they exclude the very providers who deliver most services in many settings. If they include lay providers, they must confront hard questions about what training is really necessary. The global counselling field has not yet resolved this tension, and how it does so will shape the geography of mental health care for the next generation. 7.4 Language and epistemic narrowing Language is a quieter but important dimension of homogenization. English has become the working language of the global counselling profession. Journals, conferences, and #accreditation manuals are largely in English. This has practical consequences for who can participate in the international conversation, but it also has conceptual consequences. When Chinese or Arabic or Yoruba concepts of self, family, and suffering are translated into English, they lose some of their texture. Over time, the English framework starts to seem natural, and the alternatives start to seem exotic (Mills and Fernando, 2022; Watters, 2021). For students, this raises an important question. Whose voice appears in your reading list, and whose voice does not. The answer is not neutral. 8. Counter Currents, Where the Homogenization Slows Not all forces run in the direction of similarity. Several counter currents are worth noting. 8.1 Decolonising counselling and psychotherapy A growing literature calls for the #decolonisation of counselling and psychotherapy. This work asks how the field can honor knowledge systems beyond the western canon, how training can include indigenous and community based approaches, and how power can be redistributed within the therapy room (Mills and Fernando, 2022; Ratts and Greenleaf, 2022). Some accreditation bodies have begun to incorporate this language into their standards. It remains to be seen whether the incorporation is substantive or symbolic. 8.2 Practice based evidence Alongside the push for evidence based practice, there is a growing interest in practice based evidence, in which clinicians and researchers together generate knowledge from routine practice rather than from tightly controlled trials (Barkham et al., 2021). Practice based evidence is more sensitive to local context and can capture forms of therapy that do not fit into standard trial designs. If it grows, it may partly offset the homogenizing pull of the trial based evidence hierarchy. 8.3 Digital therapies and new templates Digital #mental_health tools, from self help apps to teletherapy platforms, have created new templates that do not always fit the traditional accreditation model. Their spread has been rapid, and the regulatory response is still catching up (Torous, Bucci, Bell, Kessing, Faurholt Jepsen, Whelan, Carvalho, Keshavan, Linardon, and Firth, 2021). Digital care may either extend homogenization, by delivering the same protocol to millions of users, or disrupt it, by opening space for locally designed interventions. 8.4 Refusal at the edges At the edges of the field, some practitioners simply refuse the mainstream template. They practice under different professional titles, decline to seek accreditation, and build their own training pathways. Sometimes this reflects a principled commitment to alternative traditions. Sometimes it reflects the practical impossibility of meeting accreditation requirements in a specific setting. Either way, refusal is a reminder that institutional isomorphism, however strong, is not total. 9. Implications for Training, Supervision, and Clinical Practice 9.1 Training programs For counsellor educators, the message of the analysis is not to reject #accreditation but to hold it lightly. Standards are useful. They protect students from poor training and clients from unqualified practitioners. But standards also carry assumptions, and educators should be alert to what those assumptions include and exclude. Programs that meet the letter of a standard can still preserve local knowledge, engage indigenous scholars, and include readings that are not on the imported syllabus. Concretely, programs might consider several strategies. They can audit their reading lists for geographical and epistemic diversity. They can build partnerships with local healers and community leaders, so that students see indigenous approaches at close range. They can teach the history of the standards themselves, so that students understand where the framework comes from and why. They can invite students to compare the assumptions of the standard with the reality of their own communities. 9.2 Supervisors For supervisors, the analysis suggests a shift in stance. #Competency_based supervision is a useful tool. But supervisors should be alert to the moments when the competency framework produces feedback that reflects a cultural norm rather than a professional standard. A trainee who works in a very direct style, or a trainee whose emotional expressiveness differs from the supervisor, is not necessarily a weak clinician. Supervision that assumes a single template for the good therapist can silence trainees whose gifts do not match the template (Watkins, 2022; Bernard and Goodyear, 2021). 9.3 Clinicians in the room For clinicians, the implication is quiet but important. The way they think about clients, from the first assessment through the closing session, is shaped by templates that they did not choose. Recognizing this does not paralyse practice. It simply invites the clinician to hold the templates as tools rather than as truths, to ask what the client's own frame is, and to adapt the work accordingly. This is especially important when the client comes from a background different from the clinician. In such moments, cultural humility is not a soft optional add on. It is a core clinical skill. It requires the clinician to notice the template pulling on them, to pause it, and to make space for the client's own way of naming the trouble (Hook et al., 2022; Sue et al., 2022). 9.4 Policy makers and professional bodies For policy makers and #professional_bodies, the analysis suggests a more reflective stance toward standard setting. Standards are inevitably political. They encode a vision of what counselling is and who counsellors serve. The question is not whether to have standards but how to design them so that they are open to plural traditions, responsive to low resource settings, and revisable as knowledge grows. Recent work on adaptive #regulation offers useful models. These approaches emphasize regular review, stakeholder involvement, and explicit attention to the effects of standards on marginalized communities (Fairburn and Patel, 2022; Djelic and Sahlin Andersson, 2021). Their adoption in counselling has been partial. There is room for professional bodies to lead in this space. 10. A Research Agenda for Students This paper is written especially for students, and it closes with suggestions for research they might undertake. First, there is room for empirical mapping. Which countries have adopted which #accreditation templates, and where do the templates diverge. What is the historical path by which a specific national standard came to resemble a specific international one. Such mapping is basic but essential work. Second, there is room for ethnographic study of standards in action. What happens inside an accreditation site visit. What conversations take place when a training program adapts its curriculum to meet a standard. Which voices are heard, and which are quiet. Ethnography can reveal the political texture of standard setting that formal documents obscure. Third, there is room for comparative work on counsellor identity. How do counsellors in different countries describe what they do. Where do their descriptions converge, and where do they still diverge. Recent surveys have shown surprisingly strong convergence on the therapeutic relationship as a central value, but real divergence on other issues such as self disclosure and family involvement (Fickling et al., 2021; Corey, 2023). Fourth, there is room for work on clients. Institutional isomorphism is largely studied from the side of practitioners and educators. Clients have their own experience of it, especially when they encounter a therapy that does not fit their cultural frame. Client centred research on the reception of standardized counselling in different communities would fill a real gap. Fifth, there is room for policy oriented work. What would a genuinely pluralist accreditation system look like. How could portability agreements be designed to protect local traditions rather than efface them. How can task sharing be reconciled with the credentialing regime that dominates the profession (Patel et al., 2022). Students who take up these questions will contribute not just to scholarship but to the design of the profession they are about to join. 9.5 Students as future standard setters One more implication is worth naming. Students who train under the current standards will, in time, become the people who write the next generation of standards. They will sit on accreditation committees, chair professional associations, and lead training programs. The templates they carry with them will shape the field for another cycle. This is not a hypothetical possibility. It is a near certainty for a share of every cohort. Students who wish to influence the direction of the profession should treat the current standards not as a fixed background but as a living document that they will one day have the chance to revise. The habits of critical reading and comparative awareness that they build now are the same habits that will let them redesign the templates later. 10.5 Practical Vignettes Before turning to the discussion, it may help to make the argument concrete through short vignettes drawn from the general literature on counselling training and practice. Consider a training program in a small country that has decided to pursue international accreditation to make its graduates more mobile. To meet the standard, the program adds two courses on group counselling, expands its practicum hours, and adopts a widely used competency framework. Within three years its graduates are eligible for licensure in several neighboring countries. The program also quietly removes a locally developed course on family and community based helping because it does not map onto any competency in the imported framework. In one motion, the program has both gained portability and lost a piece of its distinctive contribution to the field. Consider a supervisor who trained abroad and returns to their home country to lead a clinical service. They introduce a competency based supervision model with weekly recorded sessions and structured feedback. The junior clinicians appreciate the clarity. Over time, however, some of them notice that the feedback focuses on behaviours drawn from the international framework, such as goal setting and outcome monitoring, while the parts of their work that draw on local traditions of listening and story sharing receive less attention. The supervisor is not hostile to local traditions. They simply have no framework in which to score them. Consider a client who comes to counselling with distress rooted in family obligations, spiritual meaning, and community expectation. The counsellor, following their training, formulates the case in terms of cognitive distortions and behavioural activation. The client complies with the interventions but experiences the therapy as slightly off centre, addressing the surface of the problem while leaving the heart of it untouched. The therapy ends with modest gains and a client who is unlikely to return. None of these vignettes are stories of failure in the usual sense. The training program is competent. The supervisor is thoughtful. The counsellor is skilled. And yet in each case something has been narrowed. That narrowing is the daily texture of institutional isomorphism. 11. Discussion The main claim of the paper is that #international_accreditation and #quality_assurance metrics are shaping #clinical_therapeutic_practices across borders in ways that go well beyond a simple sharing of good ideas. They act as coercive rules, mimetic examples, and normative templates. Together they produce a global counselling culture that is unusually consistent for a field so young. There are three broad ways to interpret this development. The first is optimistic. Convergence is the sign of a maturing profession. It reduces charlatanism, protects clients, and makes it easier for skilled counsellors to work wherever they are needed. In this view, the price of local specificity is worth paying for the gains in safety, mobility, and quality. The second is critical. Convergence hides power. It privileges the frameworks of the wealthy and the linguistically dominant. It treats therapies developed in one place as universal and treats therapies developed elsewhere as exotic. In this view, the price of homogenization is paid disproportionately by clients and clinicians in low income and non anglophone settings. The third is pragmatic. Convergence is a fact, and its effects are mixed. The task is not to celebrate it or to condemn it but to shape it. Standards can be redesigned. Accreditation manuals can be pluralized. Training can teach the history of the field it is training people to enter. The pragmatic view holds that #reflexivity, meaning the capacity of a profession to see itself and its choices clearly, is the key competence for the coming decades. This paper leans toward the pragmatic view, while acknowledging the force of the critical position. The evidence to date does not support a simple story of triumphant standardization or of unbroken domination. It supports a more subtle story in which powerful mechanisms of convergence coexist with real, if patchy, resistance. For students, the take away is not a set of conclusions but a set of habits. Read the accreditation manual carefully, and ask what it assumes. Notice which authors appear on the syllabus, and which do not. Listen to clients whose way of framing their trouble does not match your training. Take seriously the possibility that the most important part of your competence is the capacity to question the templates you have absorbed. 11.1 A note on the geography of the field It is worth pausing on the geography of the argument. The mechanisms described in this paper do not treat all countries equally. Wealthier countries, particularly those in North America, Western Europe, and Australasia, tend to set the templates. Middle income countries adapt them with varying degrees of local input. Low income countries often receive them through donor funded programs that come with their own conditions. The direction of borrowing is not symmetrical. A counselling program in the United Kingdom is unlikely to redesign itself around a template from Ghana. A counselling program in Ghana is often expected to align with a template from the United Kingdom or the United States. This asymmetry is not the result of a conspiracy. It is the outcome of longer histories of colonial education, of English as the working language of science, and of the concentration of research funding in a small number of countries. Recognising the asymmetry does not require blaming any particular institution. It does require an honest map of how influence flows in the field, and a willingness to design mechanisms that push some of the influence in the other direction (Mills and Fernando, 2022; Ratts and Greenleaf, 2022). 11.2 The temporal dimension Institutional isomorphism is also a temporal process. It unfolds over years and decades, not months. A new accreditation manual takes time to be adopted. A new textbook takes time to reach every classroom. A generation of graduates takes time to replace the one before it. This slow pace is easy to miss in day to day practice, but it matters for how the field can change. Reforms introduced today may take a full career cycle to reach the routine work of a therapist in a distant city. Students who begin to think now about the shape of the field will still be shaping it thirty years from now. 12. Limitations Several limitations should be acknowledged. First, the paper is conceptual rather than empirical. It draws on published sources rather than new data. Its claims are meant to be plausible and useful rather than definitive. Empirical work is needed to test how strongly the mechanisms described here operate in specific settings. Second, the paper draws mostly on English language literature. This is both a limitation of the author's linguistic reach and a feature of the very homogenization the paper examines. Future work in Arabic, Chinese, Spanish, French, and other languages would enrich the picture significantly. Third, the paper focuses on counselling and psychotherapy as delivered through accredited channels. It says less about informal helping, spiritual care, or community based mental health work, even though those channels serve many people. A fuller account of #mental_health provision would need to bring those channels into view. Fourth, the paper does not evaluate specific accreditation systems in detail. Different systems vary in how open they are to plural traditions, and a comparative institutional analysis would be a valuable next step. 13. Conclusion Global counselling is in the middle of a quiet transformation. Its practices are converging across borders, driven by international accreditation, cross border quality assurance, and shared training pipelines. This convergence brings real gains in safety, mobility, and #professional_identity. It also carries real costs in the marginalization of local knowledge, the flattening of clinical style, and the narrowing of the vocabulary through which suffering can be named. The concept of institutional isomorphism, first developed decades ago for organizational sociology, remains a powerful tool for understanding this transformation. Its three classical mechanisms, coercive, mimetic, and normative, all operate in the global counselling field, and their combined effect is stronger than any one alone. Recognizing this pattern is not an argument against #standards. It is an argument for reflexivity about how standards are made, who they serve, and how they might be made better. For students entering the profession now, this reflexivity is not an academic luxury. It is a clinical necessity. The client in the room does not know or care about accreditation manuals. They care whether the counsellor can meet them where they are. To meet them well, the counsellor must be able to hold the templates lightly, to question them when necessary, and to reach for what the client actually needs. That capacity, in the end, is what all the standards are for. If the standards start to obstruct it, they need to be revised. If they support it, they should be honored. The task of the next generation of counsellors is to know the difference. References Barkham, M., Lutz, W., and Castonguay, L. G. (Eds.). (2021). Bergin and Garfield's handbook of psychotherapy and behavior change (7th ed.). Wiley. Bernard, J. M., and Goodyear, R. K. (2021). Fundamentals of clinical supervision (6th ed.). Pearson. Bromley, P., and Meyer, J. W. (2021). Hyper management, neoliberal expansions of purpose and leadership. Organization Theory, 2(3), 1 to 24. Corey, G. (2023). Theory and practice of counseling and psychotherapy (11th ed.). Cengage Learning. DiMaggio, P. J., and Powell, W. W. (1983). The iron cage revisited, institutional isomorphism and collective rationality in organizational fields. American Sociological Review, 48(2), 147 to 160. Djelic, M. L., and Sahlin Andersson, K. (2021). Transnational governance, institutional dynamics of regulation. Cambridge University Press. Fairburn, C. G., and Patel, V. (2022). The impact of digital technology on psychological treatments and their dissemination. Behaviour Research and Therapy, 148, 104007. Fickling, M. J., Tangen, J. L., Graden, M. W., and Grays, D. (2021). Multicultural and social justice competence in clinical supervision. Counselor Education and Supervision, 60(1), 22 to 36. Hodges, S. (2020). The counseling practicum and internship manual, a resource for graduate counseling students (3rd ed.). Springer Publishing. Hofmann, S. G., and Hayes, S. C. (2022). The future of intervention science, process based therapy. Clinical Psychological Science, 10(1), 45 to 61. Hook, J. N., Watkins, C. E., Davis, D. E., Owen, J., Van Tongeren, D. R., and Ramos, M. J. (2022). Cultural humility in psychotherapy supervision. American Journal of Psychotherapy, 75(3), 91 to 98. Kaplan, D. M., Francis, P. C., Hermann, M. A., Baca, J. V., and Goodnough, G. E. (2021). New concepts in the counseling profession, a global perspective. Journal of Counseling and Development, 99(2), 176 to 186. Kirmayer, L. J., and Jarvis, G. E. (2022). Culturally responsive services as a path to equity in mental healthcare. Healthcare Policy, 18(2), 11 to 23. Meyer, J. W. (2022). World society, institutional theories, and the actor. Annual Review of Sociology, 48, 1 to 20. Mills, C., and Fernando, S. (2022). Globalizing mental health or pathologizing the global south, mapping the ethics, theory and practice of global mental health. Disability and the Global South, 9(1), 1 to 15. Neukrug, E. (2022). Counseling theory and practice (3rd ed.). Cognella Academic Publishing. Patel, V., Saxena, S., and Lund, C. (2022). Transforming mental health systems globally, principles and policy recommendations. The Lancet, 400(10361), 1538 to 1559. Ratts, M. J., and Greenleaf, A. T. (2022). Multicultural and social justice counseling competencies, a leadership framework for professional school counselors. Professional School Counseling, 26(1), 1 to 12. Sommers Flanagan, J., and Sommers Flanagan, R. (2022). Counseling and psychotherapy theories in context and practice, skills, strategies, and techniques (3rd ed.). Wiley. Sue, D. W., Sue, D., Neville, H. A., and Smith, L. (2022). Counseling the culturally diverse, theory and practice (9th ed.). Wiley. Torous, J., Bucci, S., Bell, I. H., Kessing, L. V., Faurholt Jepsen, M., Whelan, P., Carvalho, A. F., Keshavan, M., Linardon, J., and Firth, J. (2021). The growing field of digital psychiatry, current evidence and the future of apps, social media, chatbots, and virtual reality. World Psychiatry, 20(3), 318 to 335. Watkins, C. E. (2022). Psychotherapy supervision research, an ever changing and continually evolving landscape. Clinical Supervisor, 41(2), 175 to 198. Watters, E. (2021). Crazy like us, the globalization of the American psyche (Revised ed.). Free Press. #counselling #global_mental_health #institutional_isomorphism #accreditation #quality_assurance #cross_border_practice #cultural_competence #cultural_humility #evidence_based_practice #decolonising_therapy #counsellor_education #supervision #professional_regulation #homogenization #standards

The rapid expansion of #international_branch_campuses across the Global South and the emerging economies of Asia, the Gulf, and Africa has reopened a long-running question in higher education studies: can a university that is legally, financially, and epistemically anchored in one country genuinely act as a #cultural_steward for the country that hosts it? This article approaches that question by treating branch campuses not only as instruments of #transnational_higher_education but also as institutions with a duty of care toward the #local_heritage of their host societies. Drawing on peer-reviewed studies published between 2021 and 2026, the discussion moves from concept to case, from the #glocalization framework and #decolonial critiques of academic mobility to concrete practices at branch campuses and joint-venture universities in Malaysia, China, the United Arab Emirates, Egypt, Indonesia, and Japan. The article argues that #curricular_integration of #indigenous_knowledge, #language_heritage, and #place_based learning is not a cosmetic addition but a structural condition for the long-term legitimacy of international branches. It proposes a working model of #cultural_stewardship built on four pillars: contextual curriculum design, reciprocal community engagement, epistemic pluralism in research and teaching, and governance structures that share authority with host-country stakeholders. The article closes with implications for policy, faculty development, and student learning, and identifies gaps that student researchers can pursue. Keywords: international branch campuses; cultural heritage; curriculum internationalization; glocalization; indigenous knowledge; higher education governance 1. Introduction Higher education has always carried a double mandate. On one hand, universities push knowledge outward across borders, disciplines, and generations. On the other, they carry the memory of the societies that fund and depend on them: their languages, ceremonies, sciences, and ways of reading the world. When a university sets up a full-degree campus in another country, both mandates travel with it, and both are put under stress. The branch is expected to deliver the parent institution's academic brand, and at the same time to belong somewhere new. This paper examines that tension. It asks how #international_branch_campuses, together with joint-venture universities and other forms of #transnational_higher_education, can move beyond a franchise model of teaching and act as active stewards of the #local_heritage they are embedded in. Interest in this question has grown for three reasons. First, the number of cross-border campuses has risen sharply since the early 2000s, and the geography of expansion has shifted from a few flagship projects in the Gulf toward a much wider footprint across Southeast Asia, mainland China, Central Asia, and Sub-Saharan Africa (Liu, 2025; Birtles, 2026). Second, host-country governments are increasingly explicit that they want more than tuition revenue and rankings prestige; they want branch campuses to contribute to national knowledge systems and to the preservation of #cultural_identity (Abdou, 2025; Karim, 2025). Third, scholars working within decolonial and postcolonial traditions have turned a critical eye on the branch-campus model, arguing that unless a deliberate effort is made to integrate local content, these institutions risk reproducing forms of academic neocolonialism that quietly displace the very heritages they claim to celebrate (Jing, Ghosh, & Liu, 2022; Xu, 2021). For a student audience preparing to work inside or alongside these institutions, this debate is not abstract. It shapes what a syllabus looks like, whose thinkers appear on a reading list, which languages of instruction are permitted, and which forms of assessment count as legitimate scholarship. It also shapes how graduates think about the relationship between global competence and rootedness in place. The purpose of this article is to give students a peer-reviewed foundation for engaging with the topic and to sketch a model of #cultural_stewardship that they can apply, test, and criticize in their own contexts. A further reason for taking this discussion seriously is that the model of the international branch is not stable. The first wave of campuses, dominated by Anglophone providers moving into the Gulf and East Asia, has been joined by a second wave that includes Chinese, Japanese, and Malaysian universities operating outside their home borders (Birtles, 2026; Xu, 2025). A third wave, driven by digital delivery and joint-degree partnerships, is beginning to blur the boundary between a physical branch and a networked program (Liang, 2025; Davoodi, 2024). Each wave changes what stewardship means in practice, and each brings new host contexts, each with its own heritage, into the conversation. This article draws its evidence from peer-reviewed studies published between 2021 and 2026. The evidence base is uneven. It is strongest for Malaysia, mainland China, and Hong Kong, where a dense literature examines branch operations and joint ventures in detail. It is moderately developed for the Gulf, Indonesia, and Japan. It is thin, though growing, for Sub-Saharan Africa, Central Asia, and parts of Latin America. Where the evidence is thin, the article notes the limitation rather than filling gaps with speculation. The article proceeds in eleven parts. Section 2 clarifies terms. Section 3 sets out the historical and policy backdrop. Section 4 develops the conceptual framework of cultural stewardship. Sections 5 and 6 examine mechanisms and cases. Section 7 discusses tensions and risks. Sections 8 and 9 propose a working model and its implications. Section 10 identifies gaps and a research agenda for student projects, followed by a conclusion. 2. Defining the Terms Precise language matters because the debate around branch campuses has accumulated overlapping vocabulary. This section fixes definitions used throughout the article. An #international_branch_campus (IBC) is an entity owned, at least in part, by a foreign higher-education provider, operating in another country under the provider's name and awarding at least one qualification of the parent institution (Wilkins, Balakrishnan, & Ramasamy, 2023; Xu, 2025). This definition excludes short courses, franchised programs delivered by a local partner without a physical parent presence, and study-abroad centers that serve only outbound students. #Transnational_higher_education (TNHE) is a broader umbrella that also includes joint-venture universities, twinning arrangements, and franchised curricula (Liu, 2025; Wang & Zhang, 2024). Some Sino-foreign cooperative universities in China are best classified as #joint_ventures rather than pure branches, since they are chartered under Chinese law and jointly governed (Miani, 2025). #Cultural_heritage in the sense used here follows a broad UNESCO-inspired definition. It includes tangible expressions such as buildings, artefacts, and manuscripts, and intangible expressions such as languages, oral traditions, performing arts, rituals, and knowledge systems about the environment (Li, 2024; Nikiforova, Oktyabrskaya, Samushkina, & Yakovleva, 2026). #Indigenous_knowledge is a narrower category referring to knowledge systems developed by peoples with long historical continuity in a specific territory, often outside the formal academic canon (Kumar & Thakur, 2026; Mbah, Clifton, & Kushnir, 2024). To act as a #cultural_steward, in this article, means to accept an active duty of care toward that heritage: to protect it from erasure, to represent it accurately in teaching and research, to allow it to shape institutional practices, and to return value to the communities that carry it. Stewardship is stronger than cultural awareness and stronger than tokenistic inclusion; it entails a governance commitment. #Glocalization is the process by which global forms are reshaped in interaction with local contexts, producing hybrid outcomes rather than pure convergence (Wang, 2025; Miani, 2025). #Decolonization of curricula is the effort to identify and unwind the ways in which colonial patterns of knowledge production continue to shape what is taught, by whom, and in whose language (Heleta & Chasi, 2023; Makokotlela & Mkhasibe, 2025). Finally, #curriculum_internationalization refers to the deliberate integration of intercultural and international dimensions into the content, delivery, and outcomes of study programs (Clarke, Yang, & Harmon, 2022; Binkley, Lauridsen, & Dennis, 2025). It should not be confused with student mobility, which is only one input to internationalization. One more term deserves clarification. A hybrid institution is one whose governance, curriculum, staffing, and student body are shaped by more than one national or cultural tradition, in a way that cannot be reduced to either origin (Wang & Zhang, 2024; Xu, 2025). Most successful branch campuses are hybrid institutions in this sense, even when their marketing materials suggest otherwise. Recognizing hybridity as a normal, rather than exceptional, feature of transnational higher education matters because it shifts the analytical question from purity to composition: what elements come from where, and how are they combined? 3. Historical and Policy Backdrop The modern #branch_campus is a recent phenomenon, but its intellectual ancestry stretches back to the colonial era, when metropolitan universities established outposts across Africa, Asia, and the Americas to educate elites who could serve imperial administrations. Contemporary IBCs are legally and rhetorically different, yet the shadow of that ancestry has not fully lifted. Studies of Canadian branch operations in China and Kuwait, for example, argue that even well-intentioned export models risk reproducing #Western_hegemony through curricular content and pedagogical style, unless localizing practices are built in from the design stage (Jing, Ghosh, & Liu, 2022). The current wave of expansion was driven by a mix of pressures. From the supply side, universities in the United States, the United Kingdom, Australia, and, more recently, China and Japan, sought new revenue streams as domestic public funding tightened and as demographic contraction narrowed home enrolments (Liu, 2025; Birtles, 2026). From the demand side, host-country governments in the Gulf, in Southeast Asia, and in East Asia used education hubs as instruments of economic diversification and soft power, offering land, tax breaks, and regulatory support in exchange for name recognition and skill formation (Xu, 2025; Yang, 2022). Two policy shifts are important for understanding the cultural-stewardship question. The first is the growing insistence by host regulators that branch campuses do more than deliver a foreign degree. In China, cooperative-university regulations require Chinese and foreign partners to share governance and to teach mandatory content on Chinese language, history, and society (Miani, 2025; Liu, 2025). In Malaysia, quality assurance instruments require IBCs to demonstrate #local_relevance in program design (Xu, 2025). In the United Arab Emirates and Egypt, policy discussions increasingly frame the IBC as a bearer of national values rather than a neutral service provider (Alhaj Bakkar, Alahmad, & Abdullatif, 2026; Abdou, 2025). The second shift is intellectual. The decolonial turn in higher-education scholarship has moved from the periphery of the field to a mainstream reference point, obliging IBC administrators, faculty, and students to consider how syllabi, reading lists, assessment methods, and even the language of instruction reproduce or challenge patterns inherited from the colonial period (Heleta & Chasi, 2023; Ozyonum, 2023). This shift matters because it changes the criteria against which an international branch is judged. It also broadens the range of stakeholders whose views are treated as legitimate inputs: not only faculty and administrators, but also community leaders, heritage practitioners, and student representatives whose knowledge of place gives them standing to speak about curriculum. A third, less-discussed shift is technological. Digital tools now allow branch campuses to record, archive, and teach with heritage materials at a scale that was not possible ten years ago. Universities in the Republic of Sakha (Yakutia) have used digital archives, artificial-intelligence-assisted transcription, and virtual reconstruction to preserve and transmit tangible and intangible heritage of the region, embedding these tools in practice-oriented degree programs (Nikiforova et al., 2026). Studies of digital humanities frameworks for cross-civilizational education argue that similar approaches can support dialogue across traditions without collapsing them into a single frame (Liang, 2025). These technological affordances do not by themselves produce stewardship; they can as easily support cultural flattening. But they change what is possible. 4. Conceptual Framework: Higher Education as Cultural Steward The framework used in this article rests on four claims, each supported in the recent literature and each contestable. 4.1 The legitimacy of an IBC is partly cultural An IBC that delivers academically strong programs but fails to earn cultural legitimacy in the host society is exposed. Studies of #cultural_discount in Malaysian IBCs show that when a program's cultural content is perceived as foreign or irrelevant, students discount its value, even if the technical quality is high (Xu, 2025). Legitimacy is therefore not only a function of accreditation but also of the perceived alignment between the campus and the community that hosts it. Legitimacy operates through multiple audiences: regulators, employers, families, and cultural authorities each apply their own standards, and a campus that satisfies one but not the others sits in a fragile position. In host contexts where public opinion of foreign education providers is mixed, small perceived slights to local heritage can accumulate into serious reputational damage over years, in ways that student satisfaction surveys will not detect until it is late. 4.2 Stewardship is heavier than integration Integration can be minimal: a token module on host-country history, a food-and-festivals week, a bilingual signage policy. Stewardship, in the sense proposed here, requires that the institution accept obligations toward the maintenance of intangible heritage over time. This includes support for the languages, arts, and knowledge holders of the host community, in ways that go beyond marketing (Damopolii, Wantu, Wantu, Umar, & Obie, 2022; Nikiforova et al., 2026). Stewardship also implies a certain humility. The institution acknowledges that it is not the sole author of the culture it engages, and that community stakeholders retain authority over how their heritage is represented. This is a different posture from the more familiar one in which the university treats itself as the primary knowledge producer and community partners as sources of raw material for research. 4.3 Epistemic pluralism is a structural condition, not a slogan For stewardship to be more than rhetoric, the institution must accept knowledge systems from the host context as legitimate objects of research and teaching, not just as illustrations of foreign theory. This is where the branch-campus debate meets the wider decolonizing-curriculum literature. Scholars from South Africa, Uganda, Zambia, and India have argued that a curriculum that treats indigenous ways of knowing as ornamentation, rather than as generative epistemologies, cannot properly serve the communities in which it operates (Gqibitole, 2026; Mpuangnan & Kaisara, 2025; Makokotlela & Mkhasibe, 2025). Epistemic pluralism does not require that every knowledge tradition be treated as equally applicable to every question; it requires that traditions be engaged on their own terms and that the criteria for evaluating them be discussed openly rather than assumed. It also requires that faculty be trained to teach in this mode, which is a serious investment. 4.4 Governance must share authority The final pillar is structural. Cultural stewardship cannot be achieved by faculty goodwill alone if the governance model concentrates decisions in the parent institution. Comparative studies of NYU's campuses in Abu Dhabi and Shanghai emphasize the role of shared governance and quality-assurance systems in shaping curricular hybridity (You & Zhao, 2024). Studies of Sino-foreign cooperative universities in China and of Hong Kong intra-country branches on the mainland underline the same point: student and faculty voice, and host-country regulatory presence, are decisive for whether local content is genuinely integrated or symbolically added (Dai, 2025; Zhang & Dai, 2024). Shared governance in this sense includes formal seats for host-country representatives on academic committees, transparent processes for how curriculum decisions are made, and mechanisms by which faculty and students can contest a decision that treats local content as expendable. Together, these four pillars support a working definition of the cultural-steward IBC as an institution whose legitimacy, curriculum, epistemology, and governance are all held accountable to the #heritage of the place in which it operates. 5. Mechanisms of Curricular Integration If #cultural_stewardship is the goal, then the concrete question is how it enters the curriculum. This section identifies six mechanisms visible across the recent case literature. 5.1 Locally rooted #general_education modules Almost every branch campus offers some general-education content. The design choice that matters is whether these modules use host-country materials as their organizing spine. At Fuzhou Lakeside International School, for example, a structured experiential program uses distinctive Fujian traditions such as savory yuanxiao, Southern Fujian dragon boat activities, and the Bó Bǐng and Lào Jǐu ceremonies as core content for a cross-cultural curriculum, framed as a "demand-response-reflection" cycle rather than as a food-and-festivals veneer (Chi, 2025). Similar structured integration has been documented in Aceh, Indonesia, where universities embed #adat customary law, oral literature, and traditional ecology into modules within the national Merdeka Belajar framework (Ramli, Razali, Gadeng, Diana, & Hariadi, 2025). 5.2 #Community_engaged learning tied to place A second mechanism is the deliberate use of community-engaged and #place_based pedagogies. The Ise and Japan Study Programme, an ongoing partnership between Kogakkan University and the Ise local government, illustrates this well. Participants engage with the region's Shinto heritage through structured interaction with local stakeholders, producing scholarly interest and affinity that a classroom-only design would not generate (Teh, 2025). The Ise case is important because it shows that internationalization does not require geographic mobility; a locally rooted program can attract international attention and act as a cultural steward at the same time. 5.3 Language policy as a stewardship instrument Language is a decisive site. The choice of #English_medium_instruction at branch campuses raises well-documented concerns about linguistic hierarchies and about the marginalization of host-country languages (Hillman, Graham, & Eslami, 2021). At the same time, some IBCs use bilingual and multilingual designs to bring indigenous or minority languages into academic spaces. Studies of Yakutia's universities show how digital tools have been used to preserve and teach endangered languages of the Sakha Republic alongside Russian and English (Nikiforova et al., 2026). Studies of higher education in Gorontalo, Indonesia, show that #local_Islamic_culture can be preserved and transmitted through university language and content policies (Damopolii et al., 2022). 5.4 #Indigenous_knowledge as content, not decoration A fourth mechanism is the treatment of indigenous knowledge as a serious academic content area. A global scientometric analysis by Kumar and Thakur (2026) found that between 2016 and 2025, indigenous knowledge moved from a locally bounded topic to a globally networked field, with sharp growth after 2022. Case studies in African higher education report the integration of indigenous ecological, medical, and legal knowledge into engineering, agriculture, and law curricula, with implications for the internationalization agenda as well as for #Sustainable_Development_Goal 4.7 on cultural diversity (Mbah et al., 2024; Kikomeko, Nabbanja, & Sekiwu, 2026; Seleke, 2026). 5.5 #Faculty_development for intercultural teaching A fifth mechanism is faculty development. Even a well-designed curriculum will fail if teaching staff do not have the intercultural competence to deliver it. Binkley, Lauridsen, and Dennis (2025) argue that "local internationalization of the curriculum" depends on faculty being able to interpret cultural identity experiences in the globally diverse classroom. In IBC settings, this challenge is compounded by high turnover, mixed hiring pools, and the phenomenon that Swenddal, Nkhoma, and Gumbley (2022) call the "IBC #Othering_Loop", in which locally hired lecturers construct their campus identities as separate from their parent university, undermining integration in both directions. 5.6 Assessment that recognizes local outputs A sixth mechanism is assessment. If evaluations reward only outputs that look like those produced at the parent institution, then local knowledge, artistic production, and community engagement will be quietly downgraded. Curriculum studies from Aceh and Indonesia more broadly point to the use of reflective journals, participatory mapping, and community project reviews as alternatives to standard essay and examination formats (Ramli et al., 2025; Rahmawati, 2025). Some IBCs are experimenting with portfolio and community-defence models that let students demonstrate learning in ways legible to both academic and community audiences. 5.7 Research agendas rooted in place A seventh mechanism, less often discussed but perhaps the most durable, is the choice of research agenda. Curriculum is shaped by what faculty know how to research and publish. If departmental incentives reward only research on topics that fit the parent institution's priorities, faculty will drift away from local topics regardless of what the mission statement says. Universities in Aceh, Gorontalo, and Yakutia illustrate the alternative: institutional research funding is allocated for projects that document, preserve, or reinterpret local heritage, and faculty who lead such projects are recognized within the promotion system (Ramli et al., 2025; Damopolii et al., 2022; Nikiforova et al., 2026). Over a decade, the cumulative effect of these agendas is a scholarly infrastructure that supports the curriculum from within, rather than depending on materials imported from abroad. 5.8 Extracurricular life as heritage practice An eighth mechanism operates outside the formal curriculum. Extracurricular life, from student clubs and dormitory events to residential-college traditions, is where much cultural learning happens. Studies of student experience in transnational settings show that programs designed around local festivals, cuisine, and civic events shape the way students identify with their host place (Chi, 2025; Teh, 2025). At the same time, extracurricular design without curriculum grounding can slide into the "food and festivals" pattern that Chi (2025) warns against. The strongest examples combine structured formal learning with well-supported extracurricular engagement, so that students encounter heritage in more than one register. 6. Cases and Comparisons This section grounds the mechanisms in comparative cases, drawing on peer-reviewed literature from the last five years. The intent is not to produce a ranking but to illustrate the range of practices, and to help students see how the four-pillar framework applies. 6.1 Malaysia: cultural discount and strategic adaptation Malaysia hosts one of the world's most mature IBC ecosystems, including campuses of Nottingham, Monash, Reading, Heriot-Watt, Xiamen, and now the University of Tsukuba (Birtles, 2026; Xu, 2025). A qualitative study of two established IBCs identified two distinct adaptation pathways: content integration, in which global curricula are adapted to local contexts through region-specific disciplines while maintaining international standards, and access facilitation, in which global content is preserved and delivered through cultural-support mechanisms (Xu, 2025). Both pathways develop hybrid governance structures that derive legitimacy from multiple stakeholder groups simultaneously. The lesson for stewardship is that cultural adaptation need not weaken academic identity; carefully designed, it can produce a legitimacy amplification that reduces cultural discount and sustains long-term operation (Xu, 2025). 6.2 China: glocal tensions in cooperative universities Sino-foreign cooperative universities are formally #joint_ventures under Chinese law, and this legal status shapes how cultural content enters the curriculum. Miani's (2025) case study of a foundation year at a Sino-foreign cooperative university shows that a foundation year in China is not a simple replication of a similar UK format. Tensions emerge around the use of English as the medium of instruction, attendance monitoring, the reception of #equity_diversity_inclusion ideals, and the development of #artificial_intelligence policies. Individual agency and creativity mediate these tensions within local arrangements. In parallel, Liu (2025) shows that Chinese TNHE policy requires flexibility and adaptability from partnering institutions and a focus on mutually reciprocal relationships built on respect and trust. Studies of the Greater Bay Area report similar patterns in Hong Kong universities' intra-country branches in mainland China, where mainland Chinese students negotiate identity between global expectations and local realities (Zhang & Dai, 2024; Dai, 2025). A comparative study of student identity in IBCs versus Sino-foreign institutes found meaningful differences in how students perceive themselves as members of a foreign campus versus a locally chartered joint venture (Cai, 2024). 6.3 The United Arab Emirates: values and family in the branch-campus society The UAE has invested substantially in becoming an #education_hub. A recent mixed-methods field study of families in the UAE higher-education community, involving 208 survey respondents and 13 in-depth interviews, argues that families remain the cornerstone of preserving values and cultural identity, and that globalization, particularly through social media and globalized education, threatens familial cohesion (Alhaj Bakkar et al., 2026). The study frames globalization as both a disruptor and a potential enabler of hybrid cultural frameworks and calls for policy interventions and collaborative efforts across families, educators, and policymakers. For IBCs operating in the UAE, this framing implies that stewardship extends beyond the classroom. Course design, campus events, dormitory life, and student services policies all become part of the cultural interface between the institution and the community. 6.4 Egypt: managing polarity between globalization and decolonization Abdou (2025) proposes a #polarity_management framework for IBCs in Egypt, arguing that IBCs offer high-quality education but may inadvertently perpetuate colonial legacies by favouring Western educational models over local knowledge systems. The framework treats full autonomy for the branch and local guardrails to preserve cultural and educational integrity as two poles that must be managed together, not as a choice between them. This framing is useful because it accepts that both poles have legitimate value and shifts the question from "which side wins" to "how are we managing the tension right now". 6.5 Indonesia: local knowledge as national curriculum reform Indonesia's #Merdeka_Belajar framework has opened space for universities to embed local cultural, ecological, and philosophical values in the curriculum (Ramli et al., 2025). Qualitative work in Aceh describes the integration of adat customary law, oral literature, and traditional ecology through thematic modules and community-based assessment. Key constraints include limited institutional guidelines, inadequate pedagogical training, and accreditation systems that reward standardized indicators over local relevance. Faculty autonomy, leadership support, and community collaboration have nevertheless enabled innovative practices. Studies of Islamic higher education in Gorontalo emphasize that the university has a duty to sustain the older local genre of Islamic culture, and warn that governance is decisive for whether that duty is met (Damopolii et al., 2022). 6.6 Japan: the Tsukuba Malaysia venture Birtles (2026) analyses the University of Tsukuba's Malaysia venture, Japan's inaugural international branch campus. The literature-based discussion identifies five factors: financial motivations, the strategy of delivering a Japanese higher-education experience, leadership challenges, operational and academic factors, and the dichotomy of global integration versus local responsiveness. It suggests that the challenges of cultural integration require a nuanced approach that balances home and local contexts. As a first-of-its-kind for Japanese higher education, the venture illustrates that IBC expansion is no longer a phenomenon led only by Anglophone universities. 6.7 India and the Indian Knowledge Systems agenda India's higher-education reforms provide another instructive comparison. Curriculum design that draws on Indian Knowledge Systems, including classical texts, ethical traditions, and applied sciences, has been proposed as a means to align Indian universities with international standards without severing them from national traditions (Rathod, 2025; Joshi, 2024). Studies of decolonization efforts in Indian curricula treat this as an opportunity to reimagine the relationship between global academic standards and inherited knowledge, though critics warn of the risk of political capture and of a narrow definition of what counts as authentically Indian (Mpuangnan & Kaisara, 2025). For branch campuses operating in India, or partnering with Indian institutions, these debates cannot be sidestepped. 6.8 New York University: comparative curriculum design You and Zhao's (2024) comparative study of the liberal arts curricula at two of NYU's international campuses shows that curricula are similar in placing liberal arts at the core but vary slightly in proportion and implementation. International learning, the quality-assurance system of the host country, and local relevance combine to shape curriculum design. The importance of governance structures and internal organisational structures in the running of international campuses is emphasized. For the stewardship question, the NYU cases show that a shared brand does not require identical curricula and that host-country regulation is a genuine driver of local content. 7. Tensions, Risks, and Critiques An honest treatment of the topic must confront the risks that specialists in the field have identified. 7.1 The neocolonialism critique The most sustained critique of IBCs is that they can reproduce academic colonialism even when their stated intent is the opposite. Studies of two Canadian IBCs found that both institutions did modernize and internationalize post-secondary education in their host countries, and that they resisted neocolonial harm and localized global standards, but the risk of reproducing Western hegemony remains a structural feature of the model (Jing et al., 2022). Similar concerns are raised in analyses of IBCs in China, where the risk of epistemic dependency on the parent institution's knowledge base has been documented (Xu, 2021). 7.2 The #Othering_Loop Swenddal et al.'s (2022) grounded-theory study of lecturers at four Australian branch campuses in Southeast Asia identifies a self-reinforcing othering loop. Branch-campus lecturers often construct their individual and campus identities as separate from the wider university, and this separation undermines global integration. The finding is important because it complicates the intuition that hiring local staff automatically produces cultural rooting. Local hiring can produce local rooting only if the institution invests in identity work at the level of departments, teams, and shared research agendas. 7.3 #Cultural_discount and market pressure The idea of cultural discount, imported from media studies into higher-education analysis, describes the loss of value that occurs when cultural content produced in one context is transferred to another (Xu, 2025). Its application to IBCs is a useful reminder that even a well-intentioned curriculum can lose meaning in transit, and that stewardship requires diagnostic tools to identify where discount is occurring and why. 7.4 Language and epistemic violence The choice of English as the language of instruction is often presented as a neutral practical decision, but it carries deep implications. The concentration of academic capital in English-language traditions can crowd out local languages, disciplines, and thinkers (Hillman et al., 2021). Studies of intercultural philosophy in higher-education internationalization argue for an "epistemological polylogue" in which multiple traditions of thought engage on equal terms, rather than a monologue in one language and one canon (R'boul, 2022). 7.5 Governance capture A further risk is that governance structures can be captured either by the parent institution, producing homogeneous programs across all sites, or by narrow interests in the host country, producing programs that use the branch as a vehicle for political goals. Studies of NYU's international campuses show that curriculum outcomes are shaped by a combination of the host quality-assurance system, local relevance, and internal organisational structures, and that no single actor can determine outcomes alone (You & Zhao, 2024). The design lesson is that stewardship depends on plural governance, not on any one guardian. 7.6 Cultural essentialism and its opposite A subtler risk is that efforts to preserve local heritage can slide into cultural essentialism, treating a living, changing tradition as a fixed object to be protected. This is a common critique of curriculum reforms that identify a single canonical version of a tradition and teach it as the authentic one. Studies of curriculum reform in Indonesia and Turkey warn that heritage-based education must remain open to internal disagreement and to change over time, or it becomes a form of cultural taxidermy (Rahmawati, 2025; Alatli & Karadag, 2022). The opposite risk is cultural relativism so strong that no shared academic standards can be maintained. Stewardship must navigate between these extremes, treating heritage as living and contested while still holding academic programs accountable to defensible standards of evidence and argument. 7.7 The tension between speed and depth A final risk is temporal. Branch campuses often operate under strong pressure to demonstrate results quickly, in the form of enrolments, rankings movement, and revenue. Stewardship of heritage is slow work, measured in relationships built and materials assembled over years. When the two timelines collide, stewardship tends to lose. Institutions that succeed over the long term tend to protect at least some stewardship activities from short-term performance metrics, giving them a separate budget line and a longer evaluation cycle (Damopolii et al., 2022; Hickey & Davies, 2022). 8. Toward a Model of Cultural Stewardship Bringing the framework and the cases together, this section proposes a working model for #cultural_stewardship at international branch campuses. The model is expressed as four commitments and eight practices, each testable in student research projects. 8.1 Four commitments Commitment 1: Contextual curriculum design. The institution commits that every academic program will treat the host context as a serious source of content, not as background. Programs will be reviewed periodically for signs of #cultural_discount and revised where needed (Xu, 2025). Commitment 2: Reciprocal #community_engagement. The institution commits to relationships with communities that carry the region's heritage. These relationships are formal, long-term, and reciprocal, not extractive. The Ise and Japan Study Programme is a working example (Teh, 2025). Commitment 3: #Epistemic_pluralism in teaching and research. The institution accepts #indigenous_knowledge and other host-context knowledge systems as legitimate objects of research, teaching, and #assessment, and hires and promotes staff on that basis (Kumar & Thakur, 2026; Mbah et al., 2024). Commitment 4: Shared #governance. The institution accepts formal mechanisms through which host-country stakeholders shape decisions on curriculum, hiring, admissions, and research priorities (You & Zhao, 2024; Miani, 2025). 8.2 Eight practices Under these commitments, eight practices operationalize the model. Practice 1: Locally anchored #general_education modules. Every undergraduate program includes at least one general-education sequence built around host-country content, taught by faculty with expertise in that content (Chi, 2025; Ramli et al., 2025). Practice 2: Multilingual policies. The institution commits to at least one host-country language alongside English, with named support for endangered or minority languages where relevant (Nikiforova et al., 2026). Practice 3: Community #knowledge_partners. Cultural knowledge holders in the region are named institutional partners with formal roles in curriculum design, guest teaching, and research supervision (Gqibitole, 2026). Practice 4: #Place_based capstone projects. Final-year projects require students to engage with a local community, problem, or heritage site in a substantive way, with community feedback as part of assessment (Teh, 2025; Rahmawati, 2025). Practice 5: Structured #faculty_development. The institution runs regular training for faculty on intercultural competence, decolonial pedagogy, and local content knowledge, and links progress in these areas to promotion (Binkley et al., 2025). Practice 6: #Assessment reform. Program-level learning outcomes are revised to include locally meaningful indicators alongside international ones (Karim, 2025; Rahmawati, 2025). Practice 7: Research funding for #local_heritage. The institution allocates internal research funding for projects that document, preserve, or reinterpret local heritage, with priority given to collaborative work with community partners (Damopolii et al., 2022). Practice 8: Digital preservation infrastructure. The institution invests in digital tools for the preservation of intangible heritage, including endangered languages, oral histories, and traditional performance, following working models such as the Yakutia universities' digitalization programs (Nikiforova et al., 2026). 8.3 Applying the model The model is not a prescription. Different host contexts will emphasize different practices, and the balance will change over time. The four-pillar framework can be used as a review instrument: an institution that scores highly on curriculum but poorly on governance may be doing symbolic stewardship rather than structural stewardship, and its cultural legitimacy is unlikely to survive a crisis. The eight practices can be used as a checklist for accreditation and program review. 9. Implications for Students, Staff, and Policymakers 9.1 For students Students are not passive consumers in this system. Their choices of major, of language, of community project, and of employer send signals that institutions read. A student who insists on writing a dissertation on a local heritage topic, in a locally rooted language, using #indigenous_knowledge as a legitimate source, is doing curricular work as well as academic work. Studies of student identity in IBCs in China and Hong Kong show that #student_agency is a real force in shaping how institutions evolve (Qu, 2026; Zhang & Dai, 2024). 9.2 For academic staff For faculty, the implications are more demanding. Cultural stewardship increases the workload, at least in the short term. It requires reading beyond one's original training, building relationships with community partners, learning at least a working knowledge of a host language, and defending the legitimacy of local content in curriculum committees. Faculty studies suggest that this workload can be managed if the institution commits to structured development and to shared responsibility (Binkley et al., 2025; Al Furqani, 2024). 9.3 For institutional leaders Leaders face the question of brand risk. The intuition that a strong global brand requires uniformity across sites is powerful but not supported by the evidence. The NYU comparative study shows that variations in local proportion and implementation can be substantial without weakening the parent brand, provided that governance is well designed (You & Zhao, 2024). Leaders who resist local integration on brand grounds may be defending a form of #legitimacy that the local environment is already discounting. 9.4 For host-country policymakers Policymakers can shape outcomes at the level of regulation. The Chinese requirement that Sino-foreign cooperative universities integrate mandatory host-country content is a strong instrument, though it also raises questions of academic freedom (Liu, 2025; Miani, 2025). Softer instruments, such as quality-assurance standards that reward evidence of #community_engagement and #local_relevance, can achieve similar effects with less friction (Xu, 2025). 9.5 For parent-institution boards Parent-institution governance sits above all of this. Boards need to decide whether their branch operations are financial ventures with cultural obligations or cultural ventures with financial obligations. The choice has practical consequences. Studies of common factors in successful IBCs emphasize the need to make this choice explicitly rather than leaving it to drift (Hickey & Davies, 2022). Boards that avoid the choice tend to signal one thing publicly and reward another internally, which produces confusion at every level below them. 9.6 For community partners Community partners, including heritage practitioners, indigenous knowledge holders, museums, religious institutions, and local government bodies, are increasingly recognized as co-authors of curriculum rather than as guests. The implication for these partners is that engagement with a branch campus is now a strategic choice with real consequences for the way their traditions are represented. Studies of community-university partnerships in Uganda, South Africa, and Indonesia suggest that community partners benefit most when the terms of engagement are formalized, when compensation is clear, and when partners retain meaningful control over how their contributions are used (Kikomeko et al., 2026; Gqibitole, 2026; Ramli et al., 2025). 9.7 For accreditation bodies Accreditation and quality-assurance agencies shape what universities feel they must do. Where agencies reward only outputs that look like those of the parent institution, stewardship activities compete with more measurable priorities and lose. Where agencies build indicators for local relevance, community engagement, and epistemic pluralism into their frameworks, stewardship gains institutional protection (Xu, 2025; Karim, 2025). Some Southeast Asian and Gulf agencies have begun to move in this direction, and the effects are visible in program-review documents from IBCs in those regions. 10. Gaps and a Research Agenda for Students Several gaps in the current literature open genuine research opportunities. First, there is limited long-term evidence on whether cultural-stewardship practices actually produce measurable outcomes in terms of #heritage preservation. Most studies are qualitative and short-term. Longitudinal designs, and mixed-methods work that pairs interviews with community-level indicators of heritage vitality, are needed. Second, there is a shortage of comparative work across #host_country regions. Malaysia and China are well studied. The Gulf is moderately well studied. Sub-Saharan Africa, South Asia outside India, and Central Asia are underrepresented, even though the number of IBCs in these regions is growing (Xu, 2025; Kikomeko et al., 2026). Third, the #student_perspective is often reported through satisfaction surveys rather than through richer methods that capture how students experience the tension between global and local. Studies using critical incident analysis or participatory design would extend the field (Binkley et al., 2025). Fourth, few studies examine alumni outcomes over time. The claim that IBCs can produce graduates with both global competence and rootedness in place is plausible, but longitudinal evidence is thin. Alumni cohort studies, especially those combining interviews with career-trajectory data, would allow that claim to be tested. Fifth, the intersection of #artificial_intelligence with cultural-stewardship pedagogy is emerging but underdeveloped. Studies of digital humanities and bilingual education suggest that AI-supported tools can support the preservation of endangered languages and the interpretation of heritage materials, but they also risk producing new forms of cultural flattening (Liang, 2025; Davoodi, 2024). Finally, the model proposed here is testable. Research designs that measure institutional practices against the four commitments and eight practices, and correlate them with legitimacy, retention, and community outcomes, would produce useful evidence for policy. Sixth, the intersection between #artificial_intelligence and heritage-based pedagogy deserves careful study. AI-supported translation, image-recognition tools for manuscript work, and generative systems for language revitalization have been proposed as instruments for preserving intangible heritage. Studies of digitalization at Yakutia universities and of digital humanities frameworks for cross-civilizational dialogue suggest that these tools can be productive when used within a stewardship framework, and dangerous when used outside one (Nikiforova et al., 2026; Liang, 2025; Davoodi, 2024). Student researchers who combine technical training with cultural expertise are positioned to contribute here in ways that neither purely technical nor purely humanistic researchers can. Seventh, comparative work across institutional types would be valuable. Most existing studies focus on a single IBC, a single joint venture, or a single national context. Studies that compare, for instance, an IBC in Malaysia with a Sino-foreign cooperative university in China and a locally rooted university in Indonesia, using a shared analytical framework, would produce insights about which stewardship practices travel across contexts and which are context-specific. 11. Conclusion International branch campuses are no longer a fringe phenomenon of higher education. They are a substantial part of how knowledge, credentials, and cultural narratives move across the world. Whether they serve or erode the heritage of the places in which they operate depends on choices that are made and remade every year: which reading list, which language of instruction, which community partner, which governance seat filled or left empty. This article has argued that the strongest available response to the risks of the branch-campus model is a cultural-stewardship framework built on four commitments and eight practices. The evidence base for this framework, drawn from studies published between 2021 and 2026, is uneven but pointing consistently in the same direction. When IBCs treat host-country heritage as a serious partner rather than a marketing device, they become more legitimate, more durable, and more educationally useful for their students. When they do not, they invite the very critiques that have shadowed the branch-campus model since it began. The cases surveyed here span radically different contexts. A Sino-foreign cooperative university in the Yangtze delta and a Japanese branch campus outside Kuala Lumpur face very different regulatory environments, staff pools, and student expectations, and their approaches to cultural stewardship will look different. What they share is a structural situation. Each operates as a hybrid institution with obligations in more than one direction, and each must decide, in practice and not only in policy documents, what it owes to the heritage of its host place. For students, staff, and policymakers, the practical implication is that the debate is not settled and does not have to be lost. The tools for turning branch campuses into genuine cultural stewards exist. They require will, resources, and honest reflection on what a university is for. Students who choose to study these institutions, to critique them, to work in them, and to reform them are part of the process by which those tools become standard practice. The evidence surveyed in this article suggests that the field is ready for that generation of scholarship, and needs it. Hashtags #Higher_Education #International_Branch_Campus #Cultural_Heritage #Curriculum_Internationalization #Glocalization #Transnational_Higher_Education #Cultural_Stewardship #Indigenous_Knowledge #Decolonizing_Curriculum #Local_Heritage #Global_Academic_Curricula #Epistemic_Pluralism #Place_Based_Learning #Community_Engagement #STULIB References Abdou, Y. (2025). Between globalization and decolonization: Managing tensions in the international branch campus model. Journal of Higher Education Policy and Management. Alhaj Bakkar, Y., Alahmad, A., & Abdullatif, A. (2026). Preserving cultural identity and moral values amid globalisation: Field research within the UAE higher education. Sustainability. Al Furqani, S. (2024). Faculty constructions of internationalization. Higher Education Research and Development. Binkley, J. H., Lauridsen, K., & Dennis, J. (2025). Local internationalization of the curriculum: Opportunities for intercultural learning in the globally diverse classroom. Journal of Studies in International Education. Birtles, M. (2026). From Tsukuba to Kuala Lumpur: A literature-based discussion of Japan's first international branch campus. Higher Education Quarterly. Cai, L. (2024). Student identity in transnational higher education: International branch campuses versus Sino-foreign institutes. Compare: A Journal of Comparative and International Education. Chi, D. (2025). Bridging cultural roots and global citizenship: Integrating Fujian traditional culture at Fuzhou Lakeside International School. Journal of Contemporary Educational Research. Clarke, L., Yang, H., & Harmon, D. (2022). Internationalizing higher education curricula: Strategies and approaches. International Journal of Educational Development. Dai, K. (2025). Navigating organizational identity in transnational higher education: The case of a Hong Kong university's branch campus in mainland China. Studies in Higher Education. Damopolii, M., Wantu, H. M., Wantu, T. A., Umar, T., & Obie, M. (2022). Contribution of higher education institutions in preserving the Islamic local culture of Gorontalo, Indonesia. Journal of Ethnic and Cultural Studies. Davoodi, A. (2024). Digital echoes of heritage: Toward a culturally balanced pedagogy in technology-enhanced bilingual education. International Journal of Bilingual Education and Bilingualism. Gqibitole, K. M. (2026). Healing the curriculum: Decoloniality, Africanization and indigenization through indigenous knowledge holders. Curriculum Perspectives. Heleta, S., & Chasi, S. (2023). Curriculum decolonisation and internationalisation: A critical view from the Global South. Compare: A Journal of Comparative and International Education. Hickey, R., & Davies, S. (2022). The common factors underlying successful international branch campuses: Towards a conceptual decision-making framework. Journal of Studies in International Education. Hillman, S., Graham, K. M., & Eslami, Z. R. (2021). EMI and the international branch campus. Language Teaching Research Quarterly. Joshi, S. C. (2024). Global curriculum standards and the Indian education system: Challenges, adaptation strategies, and policy implications. Journal of International Education and Policy Studies. Alatli, H. I., & Karadag, R. (2022). The role of university in local cultural development through vernacular architectural conservation education: The case of Havran, Turkey. Journal of Cultural Heritage Management and Sustainable Development. Jing, X., Ghosh, R., & Liu, B. (2022). Resisting neocolonial harm and localising global standards at international branch campuses: A comparative study of two Canadian cases. Higher Education. Karim, A. (2025). Globalization and cultural preservation in higher education curriculum development and course design: Towards international standards and domestic requirements. EPRA International Journal of Multidisciplinary Research. Kikomeko, J., Nabbanja, C., & Sekiwu, D. (2026). Integrating indigenous knowledge transition correlates higher education curricula: A case study from Busitema University. African Journal of Higher Education Studies. Kumar, V. G. C., & Thakur, K. (2026). Indigenous knowledge and curriculum internationalization for sustainability. Sustainability. Li, S. (2024). Heritage and cultural heritage tourism: International perspectives. Journal of Heritage Tourism. Liang, Y. (2025). Cultivating civilizational dialogue: A digital humanities framework for cross-civilizational education and communication. Digital Scholarship in the Humanities. Liu, D. (2025). Transnational higher education (TNHE) in China: Policy development, student experiences, and implications. Springer. https://doi.org/10.1007/978-981-96-6313-2 Makokotlela, M., & Mkhasibe, R. (2025). Decolonisation of the curriculum through the integration of indigenous knowledge in higher education. South African Journal of Higher Education. Mbah, M., Clifton, N., & Kushnir, I. (2024). Internationalising higher education curricula for sustainable development: Considerations for indigeneity and (inter)culturality. Discover Sustainability. https://doi.org/10.1007/s43621-024-00520-y Miani, M. (2025). Glocal tensions in transnational higher education in China: The case of a foundation year in a Sino-foreign cooperative university. Higher Education Quarterly. https://doi.org/10.1111/hequ.70085 Mpuangnan, K., & Kaisara, G. (2025). Curriculum decolonization: Lessons from India. International Journal of Curriculum and Instruction. Nikiforova, V., Oktyabrskaya, I. V., Samushkina, E. V., & Yakovleva, K. M. (2026). Preservation of cultural heritage: The experience of digitalization of Yakutia universities. Sibirica: Interdisciplinary Journal of Siberian Studies. Ozyonum, E. (2023). Let's talk about decolonial internationalization. Critical Studies in Education. Qu, J. (2026). Exploring students' navigation of agency and identity formation in international branch campuses in China. Higher Education Research and Development. Rahmawati, I. (2025). Integrating local cultural values in character education: An integrative literature review within the Indonesian educational context. Journal of Character Education. Ramli, R., Razali, R., Gadeng, A. N., Diana, N., & Hariadi, J. (2025). Integrating local knowledge into higher education: A qualitative study of curriculum innovation in Aceh, Indonesia. Education Sciences, 15(9), 1214. Rathod, M. (2025). Ethics, sustainability and Indian knowledge systems: A curriculum design perspective. International Journal of Educational Development. R'boul, H. (2022). Intercultural philosophy and internationalisation of higher education: Epistemologies of the South, geopolitics of knowledge and epistemological polylogue. Studies in Higher Education. Seleke, B. (2026). Repositioning indigenous knowledge in digitally-enhanced global classrooms: A South African perspective on epistemic justice and curriculum innovation. Journal of Curriculum Studies. Swenddal, H., Nkhoma, M., & Gumbley, S. (2022). Global integration barriers at international branch campuses: The IBC othering loop. International Journal of Educational Management. Teh, H. (2025). Culture, faith and tradition as a catalyst for local sustainability: Insights from the Ise and Japan Study Programme. International Journal of Sustainability in Higher Education. Wang, S., & Zhang, Y. (2024). Transnational and transcultural: Their divergence and convergence in international higher education. Studies in Higher Education. https://doi.org/10.1080/03075079.2024.2445758 Wang, T. (2025). Toward glocal citizenship: Rethinking the traditions, goals, and curricular themes of international education. Compare: A Journal of Comparative and International Education. Wilkins, S., Balakrishnan, M. S., & Ramasamy, R. (2023). International branch campuses: The influences of country of origin and campus environment on students' institution choices and satisfaction. Journal of Marketing for Higher Education. Xu, M. (2025). How cultural discount and strategic adaptation sustain international branch campuses in Malaysia. Higher Education Policy. Xu, Z.-X. (2021). Examining neocolonialism in international branch campuses in China. Educational Philosophy and Theory. Yang, T. (2022). Identity development in institutional change of international branch campuses in Malaysia: An empirical study. Journal of Studies in International Education. You, Z., & Zhao, K. (2024). Liberal arts education in transnational higher education: A comparative study of New York University and its international campuses. Studies in Higher Education. Zhang, X., & Dai, K. (2024). Global expectations but local realities? Understanding mainland Chinese students' identity formation in Hong Kong institutions' branch campuses in the Greater Bay Area. Compare: A Journal of Comparative and International Education.

Museums are moving away from fixed, one-size-fits-all displays toward flexible digital spaces where the exhibition itself can change from one visitor to the next. This shift is being driven by a new generation of #agentic_AI systems that can plan, decide, and act with limited human supervision. In this article we look at how these autonomous #curation workflows are being used to design personalized, #cross_cultural digital exhibitions for #global_audiences. We combine ideas from museum studies, human computer interaction, machine learning, and cultural heritage informatics. We describe a layered workflow in which specialized agents handle sensing, planning, generation, personalization, and quality control. We show how large language models, knowledge graphs, and generative visual tools cooperate to select artworks, build narratives, translate cultural context, and evaluate visitor satisfaction. We also examine what happens when these systems fail: biased recommendations, cultural flattening, hallucinated facts, and loss of curator authority. Drawing on recent studies, we argue that autonomous curation is best treated as a form of collaboration between people and machines rather than a replacement for curators. Concluding sections suggest research directions in explainable curation, cross-cultural translation, ethical governance, and evaluation of long term impact. This article is written for students, researchers, and early practitioners who want a clear map of what agentic AI can and cannot do inside the digital museum. Keywords: agentic AI, museum curation, personalization, cross-cultural exhibitions, digital heritage, large language models, human-AI collaboration Introduction For most of the twentieth century, a #museum_exhibition was a physical thing. Curators chose objects, wrote wall texts, arranged rooms, and hoped the visitor would follow the path they had planned. The digital turn changed this only partly. Early online collections mostly copied the printed catalogue: long lists of objects, static images, and short captions. Visitors could browse, but the exhibition itself did not adapt to them. The situation is now shifting quickly. Museums face pressure from three directions at once. First, audiences have become used to #personalized_content in music, video, and shopping, and they expect similar treatment when they visit a cultural site online (Nie, 2025). Second, digital collections have grown so large that no human team can catalog, describe, and interpret every item in a reasonable time (Neudecker, 2022; Bratt et al., 2022). Third, cultural institutions are being asked to reach #global_audiences who speak different languages, hold different values, and read images through very different cultural lenses (Tian and Zhang, 2025; De Masi et al., 2026). At the same time, the tools available to curators have changed. Generative AI, large language models, computer vision, and recommender systems have moved from research labs into working platforms. A new step in this evolution is the rise of #agentic_AI, a term that describes software agents that can plan multi step tasks, call other tools, and act with a level of independence that traditional automation does not have (Mansilla, 2025). Instead of a single model that answers one question, an agentic system chains reasoning, retrieval, generation, and self checking into a workflow. When several agents work together on a shared goal, the result is a #multi_agent_system that can carry out complex tasks such as building an exhibition from scratch (Bijedic et al., 2026). This article studies how such workflows are used, or could be used, to autonomously curate digital exhibitions. Three ideas guide the discussion. The first is #autonomy: how much decision making can be handed to the machine before curatorial responsibility becomes unclear. The second is #personalization: how the same collection can be arranged and explained differently for each visitor without losing scholarly accuracy. The third is #cross_cultural translation: how a system trained mostly on Western data can present an African, South Asian, or Indigenous object in a way that respects the community it comes from. We ask four questions. What does a full agentic curation workflow look like from raw collection data to a live personalized exhibition? Which parts of the workflow are already working in practice, and which parts are still open research? How can such a workflow respect #cultural_diversity rather than smoothing it into a single global average? And what kinds of ethical and governance rules are needed to keep visitors, communities, and curators in the loop? To answer these questions, we review recent literature, describe a layered architecture, and discuss real prototypes and case studies. We do not propose a specific software product. Instead, we offer a conceptual map that students and early researchers can use to plan their own projects. The tone is deliberately practical. Where studies agree, we say so. Where they disagree, we point out the tension. Where the evidence is thin, we say that too. The rest of the paper is organized as follows. Section 2 reviews the literature on #digital_curation, #recommender_systems, generative AI in museums, and cross cultural heritage informatics. Section 3 introduces a conceptual framework for autonomous curation. Section 4 describes the layered agentic workflow in detail. Section 5 focuses on personalization strategies. Section 6 turns to cross cultural exhibition design. Section 7 presents short case discussions. Section 8 examines #ethical_AI and governance. Section 9 discusses limitations and future research. Section 10 closes with a summary. Literature Review The literature relevant to autonomous museum curation sits across four fields that have only recently begun to talk to each other. 2.1 Digital curation and museum studies The classical view of curation treats the curator as a scholar and gatekeeper. She selects objects, writes interpretive text, and decides how the visitor is guided through a space. Digital tools started as helpers to this workflow. Neudecker (2022) shows that libraries and museums now treat their collections as #cultural_heritage_as_data, which makes them ready for computational analysis but also raises new questions about representation and rights. Bratt and colleagues (2022) argue that curators in libraries, archives, and museums are actively re imagining legacy databases to advance more #equitable_AI, sometimes by cooperating with commercial tools and sometimes by resisting them. More recent work goes further. Oomen (2025) proposes that heritage institutions must become custodians, curators, and #co_creators at the same time. In this framing, the machine is one more collaborator alongside communities, artists, and researchers. Salah (2026) uses the term #algorithmic_memory to describe the new kind of authenticity that emerges when memory is stored, sorted, and displayed by algorithms. 2.2 Recommender systems and personalization Personalization inside museums has been studied for at least two decades. Early work focused on content based and collaborative filtering, using visitor profiles and simple location data to suggest objects. Recent studies apply deep learning to visitor behaviour. Nie (2025) trains a Long Short Term Memory network with a dynamic attention module (DALIR) that predicts what a visitor will want to see next and reports improvements above sixteen percent over previous baselines on both visit sequence and dwell time prediction. Li and Liu (2025) use a modified random forest with temporal weighting and path continuity penalties, reaching classification accuracy above ninety eight percent on the Rijksmuseum dataset. Shi (2025) reports similar work with an intelligent tour guide. Beyond simple recommendation, several teams try to model the whole visitor experience. Ferrato (2025) combines indoor positioning, recommendation, and personalization in a single system that runs in real museums. Panos and colleagues (2024) use spatial and interaction data to classify visitor types inside a virtual museum. Papakostas and colleagues (2024) apply fuzzy logic weights to virtual museum interfaces to smooth the user experience when preferences are uncertain. Ruchitha and colleagues (2026) present an explainable graph recurrent boosting framework that combines classification and regression to predict multi dimensional visitor satisfaction in #immersive_museums. 2.3 Generative AI and large language models in heritage The arrival of large language models and diffusion based image models has changed what is possible inside a museum workflow. Trichopoulos and colleagues (2023) fine tune a GPT class model to act as a #context_aware recommender for museums, showing that a language model can hold both content knowledge and situational information. Vasic and colleagues (2025) compare traditional ontology methods, LLM extended methods, and pure LLM methods for virtual museums, and they conclude that a combination of ontologies and LLMs is safer than trusting a language model alone. Multimodal LLMs make this even more interesting. Ferrato and colleagues (2025) study LLaVA on the AQUA artwork dataset and find that the model performs well on visual questions but struggles when contextual understanding is required, unless a selective knowledge integration approach is added. Rachabatuni and colleagues (2024) build a #context_aware_chatbot that uses multimodal LLMs to combine visual features and curator provided context. Reshetnikov and Marinescu (2025) tune multimodal models on crowdsourced captions specifically for heritage. Beyond text, generative visual tools now support curation directly. Schaerf and colleagues (2023) use CLIP and Stable Diffusion to remap the Helsinki Art Museum collection into synthetic panoramas of the city for its biennial. Zhao and colleagues (2024) present VAIDE, a virtual AI designer that generates web based 3D exhibition layouts. Fan (2025) proposes a generative AI framework for exhibition space design that mixes thematic modeling, multimodal prompt reasoning, and multi objective optimization to balance immersion, cost, and energy. Liu and Tan (2025) combine AI generated content with virtual reality to build immersive museum designs. Song (2025) works with XR exhibitions and links them to flow theory, arguing that generative AI can unlock #dynamic_narratives that fixed content cannot provide. 2.4 Cross cultural and ethical dimensions Studies of cross cultural digital heritage warn that AI systems can flatten or misrepresent non Western material. Tian and Zhang (2025) propose a generative AI framework for semantic compression and trustworthy transmission across UNESCO, Europeana, and Google Arts data in English, Chinese, Arabic, and French, reporting close to thirty five percent improvement in cross cultural transmission trust. De Masi and colleagues (2026) compare museum AI practice across Beijing, Turin, and Harvard and argue that political and epistemological context shapes how AI is used inside a museum. Mohamad Baai and Aris (2024) explore how AI can help preserve and promote global cultures. The ethics literature is now substantial. Tiribelli and colleagues (2024) analyze specific ethical risks such as authorship, authenticity, bias, and interpretation drift, and they call for context sensitive #AI_ethics in cultural heritage. Pansoni and colleagues (2023) design an ethical framework for AI in this field. Wibawa and Rusnalasari (2024) and Ghaith (2024) both examine the role of the human curator in the age of AI conservation. Trach (2025) reviews foreign discourse on the ethics of AI integration for cultural heritage. Kanthe and colleagues (2026) look at policy and ethics challenges in heritage digitization. Golovach (2026) reviews opportunities and risks of AI in preservation. Lakshmimol (2026) traces the ethical labyrinth this field now faces. Taken together, this literature shows a field that is moving fast in every direction. What is missing is a clear picture of how #agentic_workflows tie all these pieces into a single system that can act as an autonomous #curator. Conceptual Framework Before we describe the workflow itself, we need a conceptual frame. We define autonomous museum curation as a socio technical process in which a group of AI agents, together with human curators, communities, and visitors, jointly design digital exhibitions that adapt to context, personal history, and cultural background. Three layers matter. The first is the #object_layer, which contains the digitized cultural material: images, 3D scans, texts, videos, and their metadata. The second is the #knowledge_layer, which contains structured semantic descriptions such as ontologies, knowledge graphs, and cultural annotations (Felicetti et al., 2025; Vasic et al., 2025). The third is the #experience_layer, which contains the visitor facing exhibition: layout, narrative, media, and interaction. An #agentic_workflow moves data upward through these layers. Sensors and databases feed the object layer. Retrieval and reasoning components build the knowledge layer. Generative and personalization components produce the experience layer. What makes the workflow agentic rather than pipeline based is that the agents can plan across layers, decide when to loop back, and reject their own outputs when quality is low. The framework rests on four principles. First, #human_oversight: an autonomous system without checkpoints for curators, community representatives, and users is neither safe nor scholarly. Second, #explainability: every decision an agent makes should be traceable back to inputs, rules, and data (Ruchitha et al., 2026; Goel et al., 2025). Third, #cultural_sensitivity: the system must recognize that a Ming vase and a Yoruba mask carry different histories, and it must resist the temptation to treat them with a single Western lens (Tiribelli et al., 2024; Trach, 2025). Fourth, #reflexive_authenticity: as Salah (2026) argues, digital heritage is never a neutral copy of the past; it is a new artifact whose making must be visible. With this frame in place, we can now describe the workflow itself. Agentic AI Workflow Architecture A useful way to think about the agentic curation workflow is as a Sense, Think, Act, Learn (STAL) architecture, adapted for the heritage domain. Bijedic and colleagues (2026) apply exactly this pattern in a multi agent system with an Exhibition Curator Agent, a Conversational Guide Agent, and an Image Acquisition Agent, tested on 6,398 historical events across many civilizations. We generalize their design into six agents. 4.1 The Collection Agent The #collection_agent handles the object layer. It ingests images, texts, and metadata from museum databases, from digital repositories such as Europeana, and from open collections such as Google Arts and Culture. It cleans records, resolves duplicates, and links objects to canonical identifiers. When metadata is missing, it calls captioning models trained on cultural material (Reshetnikov and Marinescu, 2025) to fill in short descriptions. When a scan is missing, it flags the item for human digitization rather than inventing content. This agent is the workflow's memory. 4.2 The Knowledge Agent The #knowledge_agent builds structured relations between objects. It uses ontologies such as CIDOC CRM and produces #knowledge_graphs that connect creators, materials, techniques, and cultural contexts (Felicetti et al., 2025). It also connects heritage records to broader linked data. Because pure LLM systems tend to hallucinate, Vasic and colleagues (2025) recommend combining ontologies with language models: the ontology gives accuracy, the model gives natural language flexibility. This agent is the workflow's semantic brain. 4.3 The Curatorial Agent The #curatorial_agent is the closest analogue to a human curator. It receives a brief, either from a museum team or from an audience segment, and it drafts an exhibition proposal. This includes a theme, a shortlist of objects, a narrative arc, and a spatial plan. Yan and colleagues (2025) show that AI can analyze large collections and predict visitor preferences to construct dynamic exhibition experiences. Zhang and colleagues (2025) interviewed thirteen curators about GenAI supported design and production and found that generative tools help with theme selection, space design, topic research, artwork selection, and element production, but they also noted risks of fragmented workflows and copyright confusion. An important feature of this agent is that it must justify its choices. Sarna and colleagues (2025) build an #intelligent_curation system that scores artworks along visual, semantic, and emotional embeddings and produces #aesthetic_scores plus curatorial suggestions. Bakshi and colleagues (2025) extend such systems to art biennials and larger exhibitions. Goel and colleagues (2025) formalize the mathematical model behind adaptive art curation and stress the need for explainability and cultural sensitivity. 4.4 The Design Agent The #design_agent turns the proposal into a spatial layout. Zhao and colleagues (2024) present VAIDE, a virtual AI designer that generates 3D web based exhibition layouts by combining floor plan recognition and diffusion models. Fan (2025) proposes a generative framework that transforms curatorial intent into feasible, semantically coherent layouts, using multi objective optimization to balance accessibility, narrative consistency, immersion, cost, and energy efficiency, with a reported improvement of more than twenty five percent over traditional and deep generative baselines. Yan (2026) applies digital twin technology to museum exhibition space design and optimization. Das and colleagues (2025) discuss how generative AI reshapes event and exhibition design more broadly. The design agent is also where XR and VR enter the workflow. Liu and Tan (2025) show that combining AI generated content with virtual reality outperforms traditional approaches in content recommendation accuracy, virtual scene fidelity, and interaction convenience. Song (2025) uses Csikszentmihalyi's flow theory to argue that generative AI can extend XR exhibitions with real time information adaptation and #dynamic_narrative_generation. Hyun (2025) reviews immersive AI exhibition case studies. 4.5 The Personalization Agent The #personalization_agent adapts the exhibition to each visitor. It uses visitor profiles, behavioural traces, dwell time, gaze data, and short surveys to decide which objects to feature, which text to show first, and how deep to go. Nie (2025) shows that dynamic attention LSTM models capture how interests shift during a single visit. Li and Liu (2025) reach very high accuracy on tour recommendation. Ferrato (2025) integrates positioning, recommendation, and personalization inside real museums. Trichopoulos and colleagues (2023) fine tune GPT class models to act as #context_aware_recommenders that consider location, time of visit, and other contextual factors. Amador and colleagues (2026) place such personalization inside the wider tourism experience. Because personalization can trap visitors in a #filter_bubble, this agent must also inject diversity. Sarna and colleagues (2025) explicitly measure aesthetic diversity as a curation quality metric. Bratt and colleagues (2022) argue that equitable curation requires curators to deliberately push against algorithmic uniformity. 4.6 The Guide Agent Finally, the #guide_agent presents the exhibition and interacts with visitors in natural language. Trichopoulos (2023) reviews the potential of large language models for cultural heritage. Jiang and colleagues (2025) design chattable avatars that let visitors converse with historical persons, and they warn that LLMs are more suited to non factual dissemination unless carefully grounded. Rachabatuni and colleagues (2024) present a context aware chatbot for cultural heritage using multimodal LLMs. Cossatin and colleagues (2025) integrate LLMs into a heritage website and show that context dependent question suggestions further enhance user experience, especially for people with low curiosity scores who need help asking good questions. Ferrato and colleagues (2025) evaluate multimodal LLMs on question answering about artworks. Rabeena and Mohan (2025) improve heritage question answering with retrieval augmented generation. 4.7 Interaction and self correction What makes this a truly agentic workflow is that the agents communicate. The curatorial agent sends a draft to the design agent; the design agent may reject it as spatially infeasible; the curatorial agent revises; the personalization agent injects visitor specific weights; the guide agent tests the narrative on a simulated visitor. A #self_correcting_loop is essential because generative models make mistakes. Vasic and colleagues (2025) emphasise that relying only on LLMs is risky due to lack of domain specific knowledge, so structured ontologies must catch factual errors. Jo and colleagues (2022) already showed that reinforcement learning can be used to design museum layouts by scoring simple actions against multiple rules and iterating. The workflow is not linear. It is a #loop_of_loops with human checkpoints at critical steps. Personalization Strategies Personalization in an autonomous curation workflow is more than choosing the next object. It is a set of decisions about content, sequence, depth, and voice. In this section we review five strategies. 5.1 Preference modeling The first strategy is #preference_modeling. Explicit signals such as ratings and questionnaires are combined with implicit signals such as dwell time, path traces, and revisits. Nie (2025) captures this evolution with a dynamic attention LSTM, showing that a visitor's interests during a single session are not stable. Ferrato and colleagues (2022) proposed the META4RS approach, which uses museum emotion and tracking analysis for recommender systems. Papakostas and colleagues (2024) apply fuzzy weights to represent uncertain preferences. 5.2 Context awareness The second strategy is #context_awareness. Location, time, group size, language, prior knowledge, and even mood are all relevant. Trichopoulos and colleagues (2023) fine tune LLMs on context and user instructions so that recommendations shift with the situation. Ferrato (2025) shows that indoor positioning improves personalization by giving the system a real spatial anchor. Egbariya and colleagues (2024) use deep learning image recognition to build indoor localization systems for museum guides. 5.3 Adaptive narrative The third strategy is #adaptive_narrative. A visitor who enters an exhibition on ancient trade routes with strong prior knowledge should not be shown the same introductory text as a visitor who is new to the topic. Generative language models can rewrite captions and stories to match reading level, cultural background, and time available. Cossatin and colleagues (2025) show that adding LLM support to a cultural heritage website raises usability and helps low curiosity visitors ask better questions. Song (2025) argues that XR exhibitions can escape fixed narrative structures once generative AI is added, allowing real time adaptation. 5.4 Emotional and cognitive engagement The fourth strategy is engagement modeling. Ruchitha and colleagues (2026) build an explainable graph recurrent boosting framework that predicts visitor satisfaction along multiple dimensions from movement, gaze, interaction duration, engagement level, and environmental conditions. Panos and colleagues (2024) identify visitor types in virtual museums using spatial and interaction data. Once the system knows the visitor's engagement profile, the curatorial agent can adjust density, pace, and challenge. 5.5 Anti bubble mechanisms The fifth strategy is deliberate diversity. Personalization without safeguards leads to #echo_chambers even inside a museum. Yan and colleagues (2025) warn that AI curation risks #cultural_homogenization and loss of interpretive agency. Bratt and colleagues (2022) note that curators can act as agents of resistance who deliberately reintroduce marginalized voices. Practically, this means the personalization agent must reserve a share of exhibition space for objects that would not have been chosen by preference alone. Sarna and colleagues (2025) build aesthetic diversity metrics directly into their intelligent portfolio curation system. Personalization becomes a balance between fitting the visitor and expanding the visitor. A well designed autonomous museum leans on both sides. 5.6 Group and family personalization Most recommender research assumes a single user, but museums are often visited in groups. Families, school classes, and tourist groups create shared decisions that are hard to model. A #group_aware_personalization agent must weigh the interests of children against those of adults, balance language levels, and pick objects that give everyone something to talk about. Amador and colleagues (2026) look at personalization in the broader tourism experience, where group behaviour is common, and they show that shared preference profiles improve satisfaction across mixed groups. Panos and colleagues (2024) note that visitor types identified from spatial data often correspond to social configurations as much as individual taste. An agentic workflow can maintain both an individual profile and a group profile, and it can decide which one to prioritize at each decision point. 5.7 Temporal personalization Interests change over minutes, hours, and years. Nie (2025) captures this with dynamic attention, showing that within a single visit the salient exhibits shift as the visitor learns. Across visits, the story is different: a visitor may return to the same museum a year later with new knowledge or with a specific research question. An autonomous system should remember previous sessions with visitor consent, adapt to seasonal exhibitions, and refresh recommendations so that the same person is not shown the same rooms every time. Trichopoulos and colleagues (2023) design their LLM recommender around time of visit and other temporal factors, which is a step toward this kind of #temporal_personalization. Cross Cultural Digital Exhibitions Personalization for a single audience is difficult. Personalization across cultures is harder. In this section we examine why, and how agentic workflows can help. 6.1 The cross cultural problem Cultural heritage is not universal. A Japanese scroll and a Coptic icon carry different iconography, different reading order, and different assumptions about the viewer. A visitor from Nairobi will interpret an image of a Kikuyu mask differently from a visitor in Vienna. A system trained mostly on Western art literature will inherit Western categories, and it will apply them to material that does not fit. Trach (2025) warns that AI can cause cultural appropriation, distortion, and exploitation if the specifics of #cultural_context are ignored. The problem shows up in three ways. First, translation: object names, dates, and descriptions must move between languages without loss. Second, framing: the story around the object must reflect the community's own understanding. Third, presentation: the layout, palette, and pacing must feel familiar to the target audience. 6.2 Cross cultural workflows Tian and Zhang (2025) address the first dimension with a generative AI based #semantic_compression and #trustworthy_transmission framework. They combine a T5 language model with semantic alignment algorithms and a multi dimensional trust model, and they test on UNESCO, Europeana, and Google Arts and Culture data in English, Chinese, Arabic, and French. Their approach improves cross cultural transmission trustworthiness by about thirty five percent on average. De Masi and colleagues (2026) study Beijing, Turin, and Harvard side by side. They show that the same technology, once deployed, takes on very different meaning. In Beijing, AI serves state led heritage governance and digital sovereignty. In Turin, it acts as a mediator inside a participatory design process. At Harvard, it is used as a creative partner in immersive reconstruction. These three examples show that a truly cross cultural workflow cannot be a single system with a single interface. It must be configurable, and its outputs must be checked against local #epistemologies. Dhanraj and colleagues (2026) tackle scale. They report a virtual curation system that uses convolutional networks, natural language processing, and knowledge graphs to organize international digital art exhibitions of 15,000 works from forty seven countries. They achieve above ninety two percent curation relevance and an eighty seven percent user satisfaction score, and they claim close to ninety percent time efficiency gains over manual curation. Even if these numbers rest on a specific evaluation setup, they suggest that scale is now feasible for cross cultural curation, provided that quality is monitored. 6.3 Multilingual and multimodal delivery At the presentation layer, multimodal LLMs and retrieval augmented generation now allow #multilingual and #multimodal delivery. Rabeena and Mohan (2025) improve retrieval augmented generation for heritage. Reshetnikov and Marinescu (2025) tune multimodal LLMs on crowdsourced captions. Ferrato and colleagues (2025) evaluate multimodal LLMs on artwork question answering, and they show that a visitor model can be used to provide personalized answers. Rachabatuni and colleagues (2024) present a context aware chatbot for heritage that combines image understanding with curator provided context. The design agent then adapts the layout. Fan (2025) and Zhao and colleagues (2024) can generate different layouts for different audiences without rebuilding the exhibition from scratch. Combined with translated narratives and community reviewed captions, the same collection can produce many exhibitions. 6.4 The role of local communities Automation is not enough. Every study we reviewed insists that #community_participation must remain central. Mohamad Baai and Aris (2024) argue that AI should support the promotion of global cultures rather than replace the voices behind them. Oomen (2025) explicitly names heritage institutions as co creators alongside communities. The agentic workflow must include a step where local curators, artists, and community representatives review key decisions before an exhibition goes public. This step is not optional. It is what stops cross cultural curation from becoming another form of cultural extraction. 6.5 Data provenance and semantic linking Cross cultural curation rests on the quality of the underlying data. Felicetti and colleagues (2025) argue that fragmented and unstructured documentation must be transformed into #knowledge_graphs aligned with international standards such as CIDOC CRM. Their two AI assisted workflows extract structured knowledge from heritage science reports and enhance AI models with curated ontological knowledge. When multiple communities and institutions contribute data, provenance matters even more. A well designed collection agent should keep track of who supplied a record, when it was reviewed, and whether the community of origin has approved its use. Tian and Zhang (2025) build a multi dimensional trust model into their semantic compression framework for exactly this reason. Without traceable provenance, cross cultural exhibitions can accidentally propagate colonial errors or misattributions at scale. 6.6 Cultural adaptation of the interface Even the interface itself carries cultural assumptions. Reading direction, color meanings, and expectations around silence and interaction all vary. A cross cultural workflow should let the design agent produce not only translated content but also #culturally_adapted_layouts. Fan (2025) and Zhao and colleagues (2024) show that generative layout tools can produce many valid arrangements from the same brief. The next step is to let community feedback loops adjust these layouts, so that an exhibition on Persian miniatures feels different in navigation from an exhibition on West African textiles. Case Discussions To make the framework concrete, we now walk through short case discussions drawn from published prototypes. Each case highlights one part of the workflow. 7.1 A multi agent generative heritage system Bijedic and colleagues (2026) implement a multi agent system that classifies and generates exhibitions from a dataset of 6,398 historical events across civilizations and epochs. Three agents cooperate: an Exhibition Curator Agent that classifies content and generates exhibitions using LLMs, a Conversational Guide Agent that provides interactive visitor engagement, and an Image Acquisition Agent that enriches records with visual content. Deterministic classification is combined with LLM based analysis. The system also learns from administrator feedback, which improves classification accuracy over time. This case shows a clean example of a STAL architecture used for #autonomous_curation across cultures and epochs. 7.2 GenAI supported curator workflows Zhang and colleagues (2025) do not build an autonomous system. They interview thirteen curators and follow three in a real online exhibition production process. They find that GenAI helps curators in theme selection, space design, topic research, artwork selection, and element production, but the same tools can #stifle_critical_thinking, spread inaccuracies, and create fragmented workflows. This case is important because it shows what happens on the human side of the workflow. Curators are not passive users. They negotiate with the tools, they distrust some outputs, and they use others creatively. 7.3 XR meets generative AI Song (2025) reviews XR exhibitions between 2015 and 2025 and classifies them into five types. She uses Csikszentmihalyi's flow theory to evaluate immersion and finds that most XR exhibitions have limited context aware interaction and fixed narrative structures. She then proposes an integrated framework where generative AI supplies real time information adaptation, dynamic narrative generation, and personalized interaction based on visitors' linguistic, behavioural, and affective data. This is the design agent and personalization agent working together at the experience layer. 7.4 AI curation for a city biennial Schaerf and colleagues (2023) developed an installation for the Helsinki Art Biennial. They used CLIP to embed works from the Helsinki Art Museum collection and assigned them fictional coordinates in the city based on their similarity to real outdoor public art. They generated synthetic 360 degree panoramas with Stable Diffusion for each new location. The result was a machine curated map of Helsinki that merged the museum collection with an imagined public space. The case is important because it shows how #generative_curation can move museum content outside the museum walls into a public, digital, city scale exhibition. 7.5 Large scale cross cultural virtual exhibitions Dhanraj and colleagues (2026) describe an end to end virtual curation platform for large scale international exhibitions. They combine visual feature extraction, metadata NLP, and semantic linking via knowledge graphs to curate collections at scale. Their reported metrics show that the framework can handle exhibitions of 100 to more than 10,000 works. Deployment across three international virtual exhibitions provides an early real world signal, though the paper's exact evaluation criteria would need independent replication. 7.6 Explainable satisfaction modeling Ruchitha and colleagues (2026) present the eXGRBoost framework, which predicts visitor satisfaction in immersive digital museums using explainable boosting on top of a graph recurrent network. Multiple hybrid models are compared. Because the system exposes which behavioural features drive each prediction, curators can revise the exhibition rather than treating the recommendation engine as a black box. This case demonstrates how #explainable_AI closes the workflow, feeding evidence back to the curatorial agent. Across these cases, one lesson repeats. When the workflow includes explicit checkpoints for curators, communities, and evaluation, autonomous curation feels like a helpful assistant. When it does not, the same technology risks producing shallow or biased exhibitions. 7.7 Curator side production tools Sarna and colleagues (2025) build an intelligent curation system for digital portfolios that combines computer vision, transformer language models, and reinforcement training. Their evaluation uses precision, F1 score, aesthetic diversity, and an engagement index, and they present case studies of solo shows, multi artist exhibitions, and bespoke viewer experiences. Bakshi and colleagues (2025) extend similar ideas to art biennales. Goel and colleagues (2025) formalize the mathematics behind adaptive art curation for museum education. These works remind us that the curator side production tool is as important as the visitor side experience. If the curatorial agent cannot present its choices in a form that a working curator can trust, the workflow stalls. 7.8 Recommender systems at scale Recommender research has also matured. Ferrato and colleagues (2022) proposed the META4RS approach that tracks emotion during a museum visit. Ferrato (2025) integrates positioning, recommendation, and personalization in production settings. Egbariya and colleagues (2024) build a mobile visitor guide with deep learning image recognition for indoor localization. Taken together, these projects show that the personalization agent is no longer speculative. It can be deployed at scale, provided that data collection is done with visitor consent and clear privacy safeguards. Ethical and Governance Issues The ethics literature on AI in cultural heritage has grown quickly. In this section we group the main concerns into six themes. 8.1 Authenticity and authorship When a generative model writes a caption, designs a room, or reconstructs a lost artwork, who is the author? Tiribelli and colleagues (2024) list authenticity, authorship, subjectivity, and interpretation bias as central risks. Salah (2026) proposes reflexive authenticity as a way to keep the making visible, so that visitors know when they are looking at a machine generated interpretation. Ghaith (2024) argues that the #human_imperative must be preserved: technical brilliance cannot replace human judgment in heritage conservation. 8.2 Bias and representation Because most training data comes from Western sources and majority languages, generative and personalization models can under represent smaller cultures, minority groups, and communities without strong digital presence. Yan and colleagues (2025) warn of cultural homogenization. Bratt and colleagues (2022) show that curators sometimes resist corporate AI tools precisely to protect representation. Trach (2025) calls for greater cultural sensitivity and inclusiveness to prevent cultural appropriation, distortion, and exploitation. 8.3 Copyright and community rights Generative tools raise practical copyright questions. Zhang and colleagues (2025) report that curators worry about copyright when GenAI generates images or text based on existing works. Communities may hold customary rights over sacred images or ancestral objects that formal copyright law does not recognize. Kanthe and colleagues (2026) argue that policy frameworks must catch up with the pace of digitization. 8.4 Transparency and explainability If a system chooses which artworks to show and which to hide, visitors and curators must be able to understand why. Goel and colleagues (2025) build explainability and ethical constraints directly into their adaptive curation model. Ruchitha and colleagues (2026) integrate explainable boosting for exactly this reason. Vasic and colleagues (2025) argue that combining ontologies with LLMs increases interpretability because the ontology anchors the model's claims. 8.5 Human oversight and skill loss An autonomous system that runs without checkpoints can hollow out the profession it was meant to support. Ghaith (2024), Wibawa and Rusnalasari (2024), and Trach (2025) all argue for a human centered approach. Zhang and colleagues (2025) show that curators worry not only about accuracy but also about how their own critical thinking degrades if they rely too heavily on GenAI drafts. Smitha and colleagues (2025) discuss the specific challenges of curating AI generated artworks, where the boundary between tool and creator becomes blurred. 8.6 Governance and regulation Finally, the field needs governance. Pansoni and colleagues (2023) design an ethical framework for AI in cultural heritage. Tiribelli and colleagues (2024) match AI ethics principles to heritage tasks and identify actors called to intervene, including funders, professional bodies, and international organizations. Lakshmimol (2026) charts the ethical labyrinth that the field must navigate. Golovach (2026) surveys opportunities and risks. Kanthe and colleagues (2026) look at the specifics of policy. Bumgardner and colleagues (2023) show, from a different domain, that automated curation and AI workflow management systems in digital pathology can be built with strong governance, which offers lessons for museums. Any operational deployment of autonomous museum curation must include an ethics review, community engagement, and regular audits. 8.7 Privacy and visitor data Because personalization needs behavioural data, museums increasingly hold sensitive information about their visitors. Gaze traces, dwell time, and interaction logs can reveal cognitive style, emotional response, and even health signals. Ruchitha and colleagues (2026) model satisfaction from multi dimensional interaction data, which shows how rich these signals are. A responsible workflow must anonymize logs, minimize collection, offer meaningful consent, and let visitors delete their trace when they choose. Egbariya and colleagues (2024) note that indoor localization can improve guidance but also produces detailed movement data that must be handled carefully. Data policies should be part of the design agent's constraints, not an afterthought. 8.8 Environmental cost Large generative models and immersive rendering are computationally expensive. Fan (2025) explicitly includes energy efficiency as one of the objectives in the multi objective optimization behind exhibition space design. Yan (2026) applies digital twin technology, which enables offline optimization and reduces on site energy demand. Museums that adopt autonomous workflows should count the #environmental_cost of continuous inference against the benefits, especially when the same visit could be served with lighter tools. Sustainable curation is part of ethical curation. Discussion, Limitations, and Future Directions 9.1 What is working Across the studies we reviewed, several patterns are clear. Personalization inside museums is a solved problem at the technical level, with recommender systems now reaching very high accuracy on real datasets (Nie, 2025; Li and Liu, 2025; Ferrato, 2025). Generative visual tools can produce workable exhibition layouts and immersive scenes (Zhao et al., 2024; Fan, 2025; Liu and Tan, 2025). Multimodal LLMs can answer artwork related questions and translate content across languages (Ferrato et al., 2025; Rabeena and Mohan, 2025). Multi agent architectures can now be built with off the shelf components (Bijedic et al., 2026). 9.2 What is unresolved Several questions remain open. First, cross cultural depth. Most systems still assume Western frames. Community reviewed pipelines exist as prototypes but not as standard practice. Second, evaluation. Numbers such as satisfaction scores and prediction accuracy do not tell us whether visitors leave with better understanding or with a false sense of authority in the exhibition. Third, long term effects on curators and communities. The interview study by Zhang and colleagues (2025) hints that critical thinking may erode if generative tools are used without care. Fourth, hallucination and factual grounding, especially for chattable avatars of real historical persons (Jiang et al., 2025). 9.3 Limitations of this article This article is a conceptual review and does not report new empirical work. Our source selection is drawn from a fast moving literature, and important work has certainly been missed. Many of the papers we cite report metrics from a single dataset or a single institution; their claims should be treated as promising rather than final. Also, several sources rely on internal evaluation criteria that would need independent replication. Readers who plan to build systems based on this article should treat it as a starting map, not a specification. 9.4 A research agenda We suggest five directions. The first is #community_in_the_loop workflows in which local curators and community representatives review the outputs of curatorial and design agents at defined checkpoints. The second is #cross_cultural_benchmarks that stress test AI curation against non Western material, minority languages, and specialist domains. The third is #explainable_curation tools that expose reasoning chains at the object level so that visitors can see why an object was placed near another. The fourth is #long_term_evaluation that measures learning, curiosity, and cultural understanding across many visits, not just one. The fifth is #governance_infrastructure for museums that adopt autonomous workflows, including audit trails, community consent registers, and clear escalation paths when something goes wrong. Students can enter this field from many angles: computer science, museum studies, design, ethics, linguistics, or heritage management. The field needs all of them. Conclusion Autonomous museum curation is no longer science fiction. Agentic AI workflows can already sense collection data, reason over knowledge graphs, draft exhibition proposals, design spatial layouts, personalize experiences for individual visitors, and guide them in natural language. Multi agent systems tie these steps into loops that can plan, act, and learn. The result is a #digital_museum that adapts to who is visiting, when they visit, and where in the world they come from. However, autonomy is not the goal. The best evidence, both from technical benchmarks and from curator interviews, points toward a #human_AI_collaboration model. Curators remain the interpreters. Communities remain the source of meaning. Visitors remain the reason the exhibition exists at all. The agents do the parts of the job that machines are good at: scaling, matching, ranking, drafting, translating, and evaluating. The humans do the parts that matter most: choosing what deserves attention, listening to communities, and taking responsibility for what the exhibition says. For students at STULIB and for early researchers, three practical takeaways summarize this article. First, learn the technical stack, from recommender systems to multimodal LLMs, because these are the tools the field will use for the next decade. Second, learn the cultural and ethical vocabulary of heritage, because building models without this vocabulary is how good technology produces bad exhibitions. Third, work in mixed teams. Museum professionals, engineers, ethicists, community members, and designers must sit at the same table if #agentic_museum systems are to serve #global_audiences with honesty. If those three habits become normal, autonomous museum curation will do what all good curation does. It will help many different people meet many different objects in ways that expand their view of the world. References Amador, C. F., Bolanos Rodriguez, E., Chau, A. L., Gonzalez-Islas, J., and Olivera Cruz, M. S. (2026). Personalization of the tourism experience through machine learning techniques. International Journal of Combinatorial Optimization Problems and Informatics. Bakshi, S., Srishti, P., Ravichandran, J., Vadivelu, V. K., Saini, J., and Arora, A. (2025). Intelligent curation of art biennales and exhibitions. ShodhKosh: Journal of Visual and Performing Arts. Bijedic, N., Cusic, A., Kapetanovic, S., Buric, D., Cajdin, N., and Gutosic, A. (2026). Generative AI for cultural heritage promotion: A multi-agent system implementation. Bratt, S. E., Chu, C. M., Bilal, D., Choi, K., and Lafia, S. (2022). Artificial intelligence in data curation: How curators re-imagine legacy database systems to advance equitable AI in libraries, archives and museums. Proceedings of the Association for Information Science and Technology. Bumgardner, V. K. C. and colleagues (2023). Automated curation and AI workflow management system for digital pathology. Cossatin, A. G., Mauro, N., Ferrero, F., and Ardissono, L. (2025). Tell me more: integrating LLMs in a cultural heritage website for advanced information exploration support. Information Technology and Tourism. https://doi.org/10.1007/s40558-025-00312-8 Das, M. P., Dao, T., Jung, P., Leung, D. Y. H., and Liao, Y. (2025). Creative frontiers: The role of generative AI in transforming event and exhibition design. IASDR Conference Proceedings. De Masi, V., Jin, X., and Nanetti, A. (2026). Artificial intelligence in the museum experience: Comparative perspectives from Beijing, Turin, and Harvard. Athens Journal of History. https://doi.org/10.30958/ajhis.12-2-4 Dhanraj, F. M., Kurulekar, M., Pahariya, S., Bhagat, A., Singh, J., Anitha, M., and Patil, B. K. (2026). Virtual curation methods for organizing large-scale international digital art exhibitions. ShodhKosh: Journal of Visual and Performing Arts. Egbariya, B., Kuflik, T., and Shimshoni, I. (2024). Mobile museum visitors guide based on an indoor localization system using deep learning-based image recognition. Fan, J. (2025). The design method of generative artificial intelligence for exhibition spaces. Felicetti, A., Himmiche, A., and Somenzi, M. (2025). Knowledge graphs and artificial intelligence for the implementation of cognitive heritage digital twins. Applied Sciences, 15(18), 10061. Ferrato, A. (2025). Integrating indoor positioning, recommendation, and personalization to enhance museum visitor experiences. Proceedings of ACM UMAP. Ferrato, A., Limongelli, C., Gasparetti, F., Sansonetti, G., and Micarelli, A. (2025). Exploring the potential of multimodal large language models for question answering on artworks. Proceedings of ACM UMAP. https://doi.org/10.1145/3708319.3733648 Ferrato, A. and colleagues (2022). The META4RS proposal: Museum emotion and tracking analysis for recommender systems. Ghaith, K. (2024). AI integration in cultural heritage conservation: Ethical considerations and the human imperative. International Journal of Education and Digital Learning. Goel, P., Samrat, B., Juneja, B., Bhatt, R., Yashoda, L., and Das, S. (2025). Art curation algorithms: Machine learning in museum education. ShodhKosh: Journal of Visual and Performing Arts. Golovach, N. (2026). Artificial intelligence and digital technologies in cultural heritage preservation: Opportunities and risks. Hyun, J. (2025). Case studies of immersive exhibition utilizing AI. Jiang, Z., Qin, Y., and Finnegan, D. (2025). Chattable avatars: Using LLMs to power visitor engagement with historical persons. Jo, S. M., Bae, G. T., Song, H., Jung, S., Kang, Y., Shin, J., Choi, J., and Park, J. W. (2022). Artificial intelligence curation system for museums. Kanthe, R. U., Soorya, A. M. V., Patel, S., Wankhade, N., Barruga, M. B., and Jadhav, S. S. (2026). Ethics and policy challenges in the digitization of cultural heritage. Lakshmimol, C. P. (2026). Navigating the ethical labyrinth of AI in cultural heritage. Li, K. and Liu, H. (2025). Machine learning framework for adaptive museum tour recommendation and personalized visitor experience. Liu, Y. and Tan, L. (2025). Integrating generative AI and VR in museum design: An intelligent system for interactive engagement. Mansilla, R. (2025). Transforming media and entertainment workflows with agentic AI. Mohamad Baai, N. and Aris, A. (2024). AI and cultural heritage: Preserving and promoting global cultures through technology. Nano. Neudecker, C. (2022). Cultural heritage as data: Digital curation and artificial intelligence in libraries. Nie, S. (2025). Dynamic prediction and recommendation of museum visitors' interest based on long short-term memory network. Discover Artificial Intelligence. https://doi.org/10.1007/s44163-025-00472-w Oomen, J. (2025). New narrative visions for cultural heritage: Cultural heritage institutions as custodians, curators, and co-creators. Panos, F., Raptis, G., Katsini, C., and Katsanos, C. (2024). Towards identifying visitor types in virtual museums using spatial and interaction data. Proceedings of ACM UMAP. https://doi.org/10.1145/3631700.3664908 Pansoni, S., Tiribelli, S., Paolanti, M., Di Stefano, F., Frontoni, E., Malinverni, E., and Giovanola, B. (2023). Artificial intelligence and cultural heritage: Design and assessment of an ethical framework. ISPRS Archives, XLVIII-M-2-2023, 1149. Papakostas, C., Troussas, C., Krouska, A., Mylonas, P., and Sgouropoulou, C. (2024). Utilizing fuzzy weights for enhanced user experience in virtual museums. Rabeena, T. M. and Mohan, A. (2025). Enhanced multimodal retrieval augmented generation for cultural heritage preservation. Rachabatuni, P. K., Principi, F., Mazzanti, P., and Bertini, M. (2024). Context-aware chatbot using MLLMs for cultural heritage. Proceedings of ACM Multimedia Workshops. https://doi.org/10.1145/3625468.3652193 Reshetnikov, A. and Marinescu, M. C. (2025). Caption generation in cultural heritage: Crowdsourced data and tuning multimodal large language models. Proceedings of the LM4UC Workshop. Ruchitha, P., Gadapaka, K., Sharada, B., Sheela, M., Karthikeyan, M., and Sruthi, S. (2026). An explainable graph-recurrent boosting framework for multidimensional visitor satisfaction modelling in immersive digital museums. Salah, M. (2026). Algorithmic memory: Towards reflexive authenticity in cultural heritage. AI and Antiquity. Sarna, S., Bibhu, V., Das, D., Singh, Y., Guntaj, J., Rajagopal, S., and Deshpande, A. A. (2025). Intelligent curation of digital portfolios for artists. ShodhKosh: Journal of Visual and Performing Arts. Schaerf, L., Ballesteros Zapata, P., Bernasconi, V., Neri, I., and Negueruela del Castillo, D. (2023). AI art curation: Re-imagining the city of Helsinki in occasion of its Biennial. Shi, R. (2025). Personalized recommendation method for intelligent museum tour guides based on artificial intelligence. Smitha, S. P., Samanta, S., Gupta, S., Nagpal, M., Kapoor, P., and Ruchika (2025). Curating AI-generated artworks: Challenges and solutions. ShodhKosh: Journal of Visual and Performing Arts. Song, N. (2025). Generative AI-driven framework for extending the XR exhibition experience: Focusing on Csikszentmihalyi's flow theory. Tian, Z. and Zhang, W. (2025). Generative AI-driven semantic compression and trustworthy transmission mechanisms for cross-cultural digital heritage. Tiribelli, S., Pansoni, S., Frontoni, E., and Giovanola, B. (2024). Ethics of artificial intelligence for cultural heritage: Opportunities and challenges. IEEE Transactions on Technology and Society. https://doi.org/10.1109/tts.2024.3432407 Trach, Y. (2025). Ethical aspects of artificial intelligence integration in the field of cultural heritage: Foreign discourse. Trichopoulos, G. C. (2023). Large language models for cultural heritage. Trichopoulos, G. C., Konstantakis, M., Alexandridis, G., and Caridakis, G. (2023). Large language models as recommendation systems in museums. Electronics, 12(18), 3829. Vasic, I., Fill, H. G., Quattrini, R., and Pierdicca, R. (2025). Knowledge graphs vs. large language models: Competitors or partners in supporting virtual museums. Wibawa, B. and Rusnalasari, Z. (2024). Navigating ethics and innovation: The role of AI in cultural heritage. Harmonia. Yan, J., Wang, Y., and Liu, W. (2025). The application and challenges of artificial intelligence in contemporary art curation. Journal of Education, Humanities and Social Sciences. Yan, X. (2026). Intelligent design and optimization of museum exhibition spaces with digital twin technology. Zhang, J., Xiang, Y., Zhao, Y., Jin, X., Sun, J., Tong, X., Fan, M., and Lc, R. (2025). Pick the right thing: GenAI-supported design and production workflows of curators. Zhao, B., Chen, N., Jia, J., Feng, E., and Li, X. (2024). VAIDE: Virtual AI designer for Web3D exhibition layout creation. Hashtags #autonomous_museum_curation #agentic_AI #digital_exhibitions #personalized_museum #cross_cultural_heritage #global_audiences #museum_technology #generative_AI #large_language_models #knowledge_graphs #multi_agent_systems #cultural_heritage #immersive_museums #human_AI_collaboration #ethical_AI

In the twenty-first century, states no longer compete only through armies, trade tariffs, or nuclear arsenals. They also compete through stories, symbols, monuments, festivals, museums, cuisines, and languages. This article examines how #cultural_heritage has moved from the margins of international affairs to the center of #soft_power strategies. It argues that historical preservation is now a deliberate instrument of #diplomacy and a currency in #global_strategic_partnerships. The paper reviews the theoretical foundations of soft power, traces the institutional integration of heritage into foreign policy, and analyzes case studies from China, France, Italy, Japan, the United States, the Gulf states, and small island nations. It also considers the multilateral role of UNESCO, the politics of #restitution, the rise of #digital_heritage, and the risks of heritage being weaponized during armed conflict. The article concludes that when states protect the past honestly, share it generously, and negotiate its meanings with humility, heritage becomes one of the most durable resources in modern statecraft. When they use it to erase, dominate, or deceive, it becomes a liability. The findings are intended to help students, policy analysts, and cultural professionals understand why the preservation of #historical_memory has become a core matter of national strategy, and how the coming decades of climate stress, technological change, and geopolitical rivalry will reshape the field. Keywords: cultural heritage, soft power, cultural diplomacy, historical preservation, international relations, UNESCO, restitution, geocultural power, digital heritage, strategic partnerships 1. Introduction For most of the modern history of #international_relations, the study of power focused on hard capabilities. Analysts counted tanks, warships, missiles, and GDP. Culture, when mentioned at all, was treated as decorative. It belonged to the arts pages of newspapers, not to foreign ministries. That framing has collapsed. Today, ministries of foreign affairs in Paris, Beijing, Tokyo, Riyadh, and Washington all run large #cultural_diplomacy programs. Universities offer degrees in heritage management with an explicit focus on foreign policy. Multilateral organizations treat the preservation of monuments as a matter of peace and security. The idea that a restored temple, a returned artifact, or a jointly funded museum can shift the direction of a bilateral relationship is no longer strange. It is standard practice. This shift has been driven by several forces. First, the concept of #soft_power, formalized by Joseph Nye in the late 1980s and refined repeatedly since, gave officials a vocabulary to explain why culture matters strategically (Nye, 2021). Second, the growth of global tourism, digital media, and diaspora communities has made cultural symbols travel faster and further than ever. Third, several major powers have concluded that persuasion is often cheaper and more durable than coercion. Fourth, the rise of #identity_politics inside and between nations has made the control of #historical_narratives a matter of political survival. When one state can define what a shared past means, it can also shape what a shared future looks like. This article treats #historical_preservation not as a technical craft of stone conservation and archival cataloging, though it includes those things, but as a form of statecraft. It asks three main questions. How did heritage become integrated into diplomacy? What are the mechanisms through which it now operates? And what are the ethical, political, and practical limits of using the past as a resource for present-day influence? The argument is directed at students, but it draws on the same body of scholarship that informs professional practice. It is written in plain English, but it does not simplify the debates. Where scholars disagree, that disagreement is reported honestly. The article proceeds in nine sections. Section 2 sets out the theoretical framework, tracing soft power from Nye to the recent literature on #geocultural_power. Section 3 explains the main instruments of cultural diplomacy. Section 4 examines the multilateral architecture, with UNESCO at its center. Sections 5 through 7 present case studies of state practice. Section 8 addresses the difficult questions of #restitution, restitution politics, and the ethics of ownership. Section 9 turns to digital heritage, climate risk, and armed conflict. Section 10 concludes. 2. Theoretical Framework: From Soft Power to Geocultural Power 2.1 The Original Concept of Soft Power Joseph Nye introduced #soft_power in the late 1980s as a way to explain why the United States was still influential even when its relative economic and military lead was narrowing. He argued that a country's ability to get what it wants in world politics comes from three main resources: coercion, payment, and attraction. Attraction, or the ability to shape the preferences of others without threatening or bribing them, was what he called soft power. It drew on culture, political values, and foreign policies that others saw as legitimate. In a recent restatement of the concept, Nye (2021) emphasized that soft power is not the same as propaganda. It works only when the underlying reality is credible. A country that claims to stand for openness while jailing journalists will find that its cultural exports lose their appeal. For students it is important to see that soft power is not simply nice. It is a form of #power. It has purposes, targets, and measurable effects. It is meant to change behavior. What distinguishes it from hard power is the mechanism, not the ambition. A film festival, a scholarship program, or a heritage restoration project can serve national interests just as directly as a trade agreement. 2.2 Cultural Heritage Enters the Framework Nye's original writing gave more attention to popular culture, universities, and political ideals than to old buildings and archaeological sites. Later scholars filled that gap. They pointed out that #cultural_heritage is a particularly powerful soft-power resource for three reasons. It is emotionally charged, because it connects to identity and memory. It is materially rooted, because it is tied to specific places that people can visit. And it is often legally protected, because international law treats certain sites as belonging to all of humanity. These three features make heritage sticky. Unlike a pop song or a television series, a restored temple or an archaeological park cannot be easily replaced or forgotten. Clarke (2020) argued that #cultural_diplomacy is best understood as a set of practices in which cultural resources are mobilized to advance foreign policy objectives. Heritage sits at the core of that toolkit. It offers what Clarke called a long horizon of legitimacy. A state that funds the restoration of a monument in a partner country is making a claim about shared civilization that is harder to dismiss than a press release. 2.3 Geocultural Power More recently, Winter (2022) developed the concept of #geocultural_power to capture how large states, especially China, are combining infrastructure, heritage, and narrative to reshape the map of global influence. Geocultural power differs from earlier soft-power thinking in three ways. It is more explicitly territorial, tying claims of shared civilization to specific corridors such as the ancient Silk Roads. It is more infrastructural, using ports, railways, museums, and archaeological missions as physical anchors. And it is more strategic, being coordinated at the highest levels of the state rather than left to cultural attaches. Winter's framework is useful because it captures the scale of what is happening. When a state signs heritage cooperation agreements with dozens of countries, funds joint archaeological digs across two continents, and builds new museums to host traveling exhibitions, it is not simply doing cultural outreach. It is constructing a new #geopolitical_imagination. Students should notice how this contrasts with older debates about whether culture is really about power at all. In the geocultural view, the answer is unambiguous. Culture is one of the primary terrains of strategic competition. 2.4 Critical Voices Not all scholars accept these framings without qualification. Labadi (2022) warned that treating heritage as a soft-power resource can distort preservation practice. When ministries want quick diplomatic wins, they may push for spectacular restorations of famous monuments while ignoring less photogenic sites that matter more to local communities. #Sustainable_heritage requires participation, not just visibility. Similarly, Harrison and colleagues (2020) argued that framing heritage only in terms of national interest ignores the possibility of building shared futures across borders, and can foreclose the more radical potential of heritage to imagine alternatives to the state itself. These critical voices are essential. They remind us that a strategy can be effective on its own terms and still be ethically thin. 3. Instruments of Cultural Heritage Diplomacy If heritage is a soft-power resource, what are the actual tools through which it is deployed? This section identifies six main instruments. They overlap in practice, but separating them helps clarify how the field works. 3.1 Bilateral Preservation Agreements The most direct instrument is the bilateral #heritage_agreement. Two states sign a treaty or memorandum in which one commits to fund or provide technical support for the restoration of a monument located in the other. France has used this tool intensively in North Africa, the Middle East, and Southeast Asia. Italy has done the same across the Mediterranean and in Latin America. Japan has developed a distinctive model of low-key, long-term technical cooperation, especially for Buddhist heritage in South and Southeast Asia. The United States runs the Ambassadors Fund for Cultural Preservation, which provides small but symbolically important grants to preserve sites in more than one hundred countries. Bilateral agreements are powerful because they combine tangible investment with a public act of respect. A ribbon-cutting at a restored temple is a diplomatic performance. It signals that the donor country values the recipient's past. It creates an ongoing relationship, because restored sites need maintenance. And it produces a physical object that will stand for decades, embodying the partnership long after the officials who signed the deal have moved on. 3.2 Multilateral Frameworks The second instrument is participation in multilateral frameworks. UNESCO is the anchor institution, and its #World_Heritage system will be discussed in the next section. But the field is wider than UNESCO. There are the International Council on Monuments and Sites, the International Centre for the Study of the Preservation and Restoration of Cultural Property, and regional bodies such as the Arab League Educational, Cultural and Scientific Organization and the Association of Southeast Asian Nations cultural committees. States accumulate influence within these bodies by hosting meetings, funding programs, and placing their nationals in leadership roles. The soft-power payoff is diffuse but real. It sets the terms of what counts as good practice in preservation. 3.3 Museum Diplomacy The third instrument is the museum. #Museum_diplomacy has grown rapidly over the last two decades. Grincheva (2020) documented how national museums have become platforms for international engagement, using loans, joint exhibitions, digital projects, and staff exchanges to build ties across borders. The Louvre Abu Dhabi is the most spectacular example of this trend. Under a long-term agreement, France licensed its brand and lent works from its national collections to a new museum in the United Arab Emirates. The arrangement was framed as a partnership of civilizations, but it also produced substantial financial returns for France and reinforced Emirati ambitions to position Abu Dhabi as a cultural capital of the region. Museums matter because they concentrate objects, expertise, and audiences in one place. A traveling exhibition of ancient bronzes or medieval manuscripts can reach hundreds of thousands of visitors and generate media coverage in ways that a bilateral treaty cannot. Museums also provide credible platforms. When a respected institution presents a country's heritage, the presentation carries more weight than official communication because the institution has its own reputation to protect. 3.4 Archaeological Missions The fourth instrument is the foreign #archaeological_mission. For more than a century, European and North American universities have run digs in the Mediterranean, the Middle East, Central Asia, and Mesoamerica. These missions were originally often extractive, sending finds home to metropolitan museums. Contemporary practice is different, at least in its stated principles. Finds now stay in host countries, and missions are expected to train local scholars and build local capacity. But the diplomatic function remains. States such as France, Germany, Italy, Japan, and increasingly China maintain networks of missions that serve as durable channels of intellectual and political engagement. 3.5 Language and Living Heritage The fifth instrument is the promotion of language and #intangible_heritage. Language institutes, such as the Alliance Francaise, Goethe-Institut, Instituto Cervantes, British Council, Confucius Institute, and Yunus Emre Institute, are among the most visible expressions of state soft power. They teach language, but they also convene events on literature, music, food, and film. Living heritage, in the sense used by the UNESCO 2003 convention, extends this domain to traditional crafts, oral traditions, and ritual practices. Recognition of an element on the intangible heritage list carries prestige and can boost tourism, but it also raises delicate questions about who owns living traditions and how they should be shared. 3.6 Digital Platforms The sixth instrument, still developing, is the #digital_platform. Google Arts and Culture, the Europeana project, national digitization programs, and virtual reality reconstructions of destroyed sites are all forms of digital heritage diplomacy. Grincheva (2020) argued that digitization does not simply extend the reach of museums. It changes their character. Digital heritage is more participatory, more remixable, and more vulnerable to manipulation. States that invest in high-quality digital platforms gain reach, but they also expose themselves to new forms of contestation, including deepfake reconstructions and politically charged remixes of national symbols. Taken together, these six instruments constitute a mature toolkit. What is striking is how quickly they have been institutionalized. Twenty years ago, only a handful of countries had explicit cultural-diplomacy strategies. Today, most middle powers do. 4. UNESCO and the Multilateral Governance of Heritage Any serious discussion of heritage in international relations must engage with UNESCO. The United Nations Educational, Scientific and Cultural Organization was created in 1945 with the founding conviction, written into its constitution, that since wars begin in the minds of men, it is in the minds of men that the defenses of peace must be constructed. Culture was thus placed alongside education and science as a pillar of postwar peacebuilding. 4.1 The World Heritage System The #World_Heritage_Convention of 1972 gave the organization its most visible instrument. Sites of outstanding universal value could be inscribed on a global list, subject to protection commitments. Over five decades, the list has grown to include more than 1,100 sites in more than 160 countries. Inscription is a mixed blessing. It boosts tourism and attracts investment, but it also imposes obligations and can provoke conflict between national governments and local communities. Meskell (2020) has written extensively on the politics of the World Heritage Committee. She has shown that decisions about which sites are inscribed, which are placed on the list of endangered sites, and which are removed are shaped by voting blocs, informal deals, and strategic alliances. States use their positions on the committee to protect their own sites and to reward or punish partners. This is not corruption in a legal sense. It is politics in the ordinary sense. But it means that the World Heritage system is not a neutral technical process. It is a diplomatic arena in which reputational capital is won and lost. 4.2 Emergency Response and the Protection of Heritage in Conflict A second area of UNESCO practice is the protection of heritage during armed conflict. The 1954 Hague Convention for the Protection of Cultural Property in the Event of Armed Conflict provided the legal framework. Later protocols strengthened it. In recent years, UNESCO has mobilized international responses to the destruction of heritage in Syria, Iraq, Yemen, Mali, and Ukraine. The 2023 partial destruction of the Old City of Odesa, which had been inscribed on the World Heritage list in emergency proceedings in 2023, illustrated both the ambition and the limits of the system. Inscription can raise the reputational cost of attacks, but it cannot stop them. 4.3 Intangible Heritage and Cultural Diversity The 2003 Convention for the Safeguarding of the Intangible Cultural Heritage extended UNESCO's reach beyond monuments to include living practices. This shift was politically important. It gave voice to countries and communities whose heritage did not consist of large stone structures. It also opened new spaces for contestation. Elements are proposed by states, but they are often shared across borders. Kimchi-making, coffee culture, calligraphy, and many forms of music and dance are claimed by multiple national traditions. The 2005 Convention on the Protection and Promotion of the Diversity of Cultural Expressions further institutionalized the idea that #cultural_diversity is a public good deserving state and international support. 4.4 Structural Tensions For all its achievements, UNESCO faces structural tensions. Its budget is limited. Major powers have withdrawn and rejoined at different times, often in response to specific political disputes. Its decisions are made by state parties, which means that community voices are filtered through governments. Labadi (2022) argued that the organization's approach to heritage remains too focused on tangible monuments and too tied to national branding, and that its potential contribution to #sustainable_development requires deeper rethinking. Nevertheless, no other institution has anything like UNESCO's convening power. For states seeking to build heritage-based soft power, engagement with UNESCO is not optional. It is foundational. 5. Case Study: China and the Silk Roads No contemporary example illustrates the integration of heritage into strategic ambition more clearly than China's revival of the #Silk_Roads. Since the announcement of the Belt and Road Initiative in 2013, Chinese authorities have combined infrastructure investment, diplomatic outreach, and heritage narratives into a single strategic frame. 5.1 From Trade Route to Civilization Narrative Winter (2022) traced how the historical Silk Roads, which for centuries were understood primarily as trade networks, have been reframed as a shared #civilizational_heritage linking China with Central Asia, the Middle East, Africa, and Europe. This reframing serves several purposes. It positions China as a longstanding partner rather than a new entrant. It legitimizes infrastructure projects as reconstructions of ancient routes rather than expansions of Chinese influence. And it provides a rich vocabulary of symbols, from camels and caravanserais to porcelain and calligraphy, that can be mobilized in diplomatic settings. 5.2 Joint Heritage Projects and Nominations Under this framing, Chinese institutions have pursued joint archaeological missions, restoration projects, and World Heritage nominations across the Silk Roads. The transnational inscription of the Silk Roads: the Routes Network of Chang'an-Tianshan Corridor in 2014 was a landmark. It involved China, Kazakhstan, and Kyrgyzstan and demonstrated that heritage diplomacy could produce multilateral outcomes previously thought difficult. Since then, further transnational nominations have been prepared, and Chinese institutions have taken increasingly active roles in heritage bodies. 5.3 Museum Networks and Academic Diplomacy Chinese museums have partnered with counterparts in dozens of countries, hosting large exhibitions of Chinese artifacts and receiving reciprocal displays. Universities have created new programs in Silk Road studies, and the state has funded scholarships that bring students from partner countries to study heritage in China. These programs recruit and cultivate future professionals who will be sympathetic to Chinese framings. 5.4 Critiques Critics have raised several concerns. Some argue that the emphasis on civilizational continuity obscures episodes of conflict and coercion that were also part of Silk Road history. Others question whether local communities in partner countries genuinely benefit or whether they are primarily audiences for a narrative constructed elsewhere. There are also concerns about the sustainability of restoration projects when Chinese partners depart. These critiques do not necessarily undermine the strategy. They highlight the tensions inherent in any large-scale heritage program that is also an instrument of foreign policy. Still, the Silk Roads case demonstrates the scale that is possible. It shows how a state can combine narrative, infrastructure, and preservation into a coherent geocultural strategy that operates over decades and reshapes the terms of international engagement. 6. Case Study: European Powers and Historical Presence European states approach cultural heritage diplomacy from a different starting point. Their historical presence in the Middle East, Africa, Asia, and the Americas is bound up with colonial pasts that are increasingly contested. This creates both opportunities and constraints. 6.1 France France has one of the most developed cultural diplomacy systems in the world. The Ministry for Europe and Foreign Affairs works with the Institut Francais, the network of Alliances Francaises, and specialized bodies such as the French archaeological schools in Athens, Rome, Cairo, Jerusalem, and further afield. Together they operate a global network of #cultural_presence. In recent years, French cultural diplomacy has emphasized partnership and co-creation, in part as a response to criticism of the colonial history of French museums. The Louvre Abu Dhabi, opened in 2017, remains a central case. Under a thirty-year agreement, France licensed the Louvre name and provided loans from its national collections. The project has been described both as a bold experiment in shared universalism and as the commodification of French cultural capital. The reality is that both descriptions are partly true. The project brought hundreds of millions of dollars to French institutions, enhanced their capacity, and created a durable Franco-Emirati relationship. It also raised legitimate questions about how cultural heritage can be sold as a service. France has also engaged seriously with the question of #restitution, particularly in relation to African heritage. A 2018 report commissioned by the French presidency proposed a framework for the return of objects taken during the colonial period. Subsequent legislation and case-by-case returns have made France one of the first major European states to move systematically on this issue. Progress has been slower than initial announcements suggested, but the direction of travel is clear (Sarr and Savoy, 2018, foundational, cited in later work by Vergo, 2021, and others). For students, France offers a case in which soft power is being reconstructed under the pressure of critical scrutiny. 6.2 Italy Italy's cultural diplomacy leans heavily on its Mediterranean identity and its extraordinary density of heritage sites. It hosts more UNESCO World Heritage sites than any other country. Italian conservators are internationally recognized. The Ministry of Foreign Affairs and International Cooperation runs a network of Italian Cultural Institutes and supports archaeological missions in more than 40 countries. Italy has been particularly active in heritage rescue after natural disasters and conflicts, contributing expertise in Iraq, Syria, Nepal, and elsewhere. This positioning gives Italy a reputational asset. When cultural emergencies arise, Italian teams are often among the first called. 6.3 The United Kingdom and Germany The British Council operates in more than 100 countries. Its work covers language, education, and arts, with heritage embedded across these areas. Germany, through the Goethe-Institut and the Deutsches Archaologisches Institut, maintains one of the most respected networks of archaeological missions in the world. Both countries face intensified debate about #restitution, particularly the Benin Bronzes and other objects held in national and municipal museums. Germany's decisions to transfer legal ownership of large groups of bronzes to Nigeria in 2022 marked a significant shift, though disputes about their subsequent management have complicated the story. The soft-power lesson is that returning objects can be as powerful as displaying them, provided the returns are done honestly and are supported by long-term cooperation. 6.4 Central and Eastern European Voices Central and Eastern European states, whose heritage was often controlled or damaged during the Soviet period, have used post-1989 preservation efforts to reassert national identity and rebuild ties with Western partners. Poland, Ukraine, Czech Republic, and the Baltic states have integrated heritage into their foreign policies. The destruction of Ukrainian cultural sites since 2022 has drawn intensive international attention, and rescue and documentation efforts have become part of broader partnerships with European and North American institutions. Across Europe, the pattern is clear. Heritage remains an important instrument, but it must now be handled with an awareness of contested pasts. Uncritical celebration is no longer credible. The most effective European strategies combine investment, partnership, and honest engagement with difficult histories. 7. Case Study: Rising and Middle Powers While the major powers have long-established programs, some of the most interesting recent developments come from states that have only recently begun to invest in heritage diplomacy on a large scale. 7.1 Japan Japan's approach is distinctive for its emphasis on quiet, long-term technical cooperation. Japanese teams have worked on the restoration of Angkor Wat in Cambodia, Bamiyan in Afghanistan, and many other sites across Asia. The Japanese government links this work explicitly to peacebuilding and to the promotion of #cultural_diversity. Japanese conservators are known for their patience and their attention to intangible dimensions such as the training of local craftspeople. This approach yields long-lasting relationships and a reputation that is difficult for larger, louder actors to replicate. 7.2 South Korea South Korea has moved rapidly from being primarily an exporter of popular culture, with the Korean Wave in music, television, and film, to also becoming an active player in heritage diplomacy. Korean institutions have supported restoration projects in Southeast Asia and Central Asia and have promoted Korean traditional performing arts and crafts on international stages. Korea's success in linking popular and traditional culture within a single strategic frame is instructive. #Popular_culture and heritage reinforce each other, provided both are supported with credibility and depth. 7.3 The Gulf States The Gulf states have made some of the most ambitious cultural investments of the past two decades. The United Arab Emirates has built the Louvre Abu Dhabi and is developing further museum quarters on Saadiyat Island. Qatar has established the Museum of Islamic Art and a network of other institutions in Doha. Saudi Arabia has invested heavily in the AlUla site and its wider heritage sector as part of its Vision 2030 program. These initiatives combine tourism ambitions, urban development, and diplomatic outreach. They also raise sensitive questions about labor conditions, freedom of expression, and how selectively histories are told. For students, the Gulf cases show that heritage-based soft power requires large budgets to be sustained, but budgets alone do not produce credibility. Credibility depends on how the past is handled, not only on how the buildings look. 7.4 Turkey and Iran Turkey has developed a distinctive cultural diplomacy program through the Yunus Emre Institute and the Turkish Cooperation and Coordination Agency. Restoration projects in the Balkans, the Caucasus, and Central Asia are framed as the preservation of Ottoman heritage. This framing has strategic implications, positioning Turkey as a natural cultural partner across a wide region. Iran, despite economic constraints, has invested in the preservation of Persian heritage sites at home and in cultural outreach that emphasizes Persian civilizational reach across South and Central Asia. 7.5 India India has increased attention to cultural diplomacy through the Indian Council for Cultural Relations, yoga diplomacy, and cooperation on Buddhist heritage sites across Asia. Buddhist circuits linking India with Sri Lanka, Nepal, Myanmar, Thailand, and beyond are being developed as tourism corridors and as diplomatic assets. Given India's demographic weight and its historical role as a source of religious and philosophical traditions, its potential for #heritage_diplomacy is substantial and still being realized. 7.6 Small Island States Small island states, particularly in the Pacific and Caribbean, use heritage diplomacy in yet another way. For them, the recognition of #intangible_heritage, traditional navigation, oral traditions, and endangered languages, is often more strategic than the promotion of monuments. Heritage becomes a way to project identity onto the international stage despite small populations and limited resources. It also becomes a way to link cultural survival with the existential threat of #climate_change. When a rising sea threatens ancestral lands, heritage diplomacy and climate diplomacy become inseparable. Across all of these cases, the common thread is that heritage is being integrated into national strategy. The specific mix of tools, ambitions, and constraints differs, but the fundamental logic is shared. 8. The Politics of Restitution and Ownership No serious discussion of #cultural_heritage in international relations today can avoid the question of #restitution. Objects taken during colonial rule, during wartime looting, or through illicit trade continue to sit in museums far from where they were made. How these objects are handled has become a marker of diplomatic maturity. 8.1 The Historical Background Nineteenth and early twentieth century museum collections in Europe and North America were assembled in conditions ranging from open theft to negotiated purchase, with a great deal of coercive ambiguity in between. Sacred objects, royal regalia, human remains, and archaeological material were removed from communities that often had no meaningful choice. For decades, curators argued that these objects were safer in metropolitan museums, that they belonged to a universal patrimony, and that returning them would empty the world's great collections. These arguments have lost credibility. 8.2 The Contemporary Shift Over the past decade, and especially since 2018, a series of high-profile cases has changed the terms of debate. The French government commissioned the Sarr-Savoy report, which recommended systematic restitution of African objects held in French museums. Germany moved to transfer ownership of large numbers of Benin Bronzes to Nigeria. The Netherlands, Belgium, and other European states have followed with their own policies. Museums in the United States have returned Native American human remains and sacred objects under the Native American Graves Protection and Repatriation Act framework, which continues to be strengthened. For strategic purposes, the significance is straightforward. #Restitution is an act of diplomatic power. When a former colonial power returns objects, it changes the terms of its relationship with the source country. Done well, restitution builds trust, opens new avenues of cooperation, and demonstrates that a state can face its history honestly. Done badly, restitution can look like a public relations exercise, or worse, can end with objects returning to conditions where they are again vulnerable to loss. 8.3 Ongoing Disputes Not all disputes are colonial. The Parthenon Marbles held in the British Museum remain a subject of long-running negotiation between the United Kingdom and Greece. The dispute over ownership and display of the Amber Room and other objects displaced during World War II continues to affect Russian and Central European relationships. Wartime looting during more recent conflicts, especially in Iraq, Syria, and Yemen, has produced an urgent need for international cooperation on illicit trade. Interpol and UNESCO maintain databases of missing objects, and states pursue legal action against traffickers. 8.4 Beyond Returns Restitution debates have also encouraged deeper thinking about ownership. Some scholars argue that the question is not only who holds an object, but who has the right to interpret it. Digital returns, in which high-quality documentation and full metadata are transferred to source communities, are one response. Joint stewardship arrangements, in which objects remain in one location but are curated collaboratively, are another. #Shared_stewardship models will likely become more common as museums seek ways to maintain global collections while acknowledging the legitimate claims of communities of origin. For students, the key point is that restitution is not a technical afterthought. It is one of the central diplomatic issues of contemporary heritage practice. States that handle it with honesty and imagination gain influence. Those that resist for too long lose it. 9. Digital Heritage, Climate Risk, and Conflict The final substantive section turns to three forces that are reshaping the field: digital transformation, climate change, and armed conflict. Each raises new possibilities and new dangers. 9.1 Digital Heritage The digitization of collections, the creation of virtual reconstructions of destroyed sites, and the use of augmented reality in exhibitions have opened new frontiers for heritage diplomacy. When Palmyra was damaged by the Islamic State, three-dimensional reconstructions of its Arch of Triumph traveled to public spaces in London, New York, and elsewhere. The reconstructions were more than technical achievements. They were diplomatic statements about the refusal to accept cultural erasure. Grincheva (2020) argued that #digital_heritage changes the audience relationship. It expands reach, allows interaction, and enables participation from anywhere with internet access. It also creates risks. Digital archives can be hacked, corrupted, or manipulated. Reconstructions can privilege certain historical moments over others and can lock in interpretations that ought to remain open. High-quality metadata and open standards matter more than ever, but they are unevenly distributed across countries. Artificial intelligence brings both promise and peril. Machine learning can accelerate the analysis of large archaeological datasets, translate ancient texts, and identify patterns invisible to human researchers. It can also produce plausible but false content, including fake objects and fabricated histories. As #artificial_intelligence tools become more accessible, heritage institutions and diplomats will need to become adept at distinguishing authentic material from generated material, and at communicating this distinction to public audiences. 9.2 Climate Change Climate change is now a heritage emergency. Rising seas threaten coastal sites from Alexandria to Venice to the low-lying atolls of the Pacific. Melting permafrost is releasing archaeological material in Siberia and destabilizing sites in the Arctic. Wildfires damage sites in the Mediterranean and Australia. Increased humidity and temperature swings accelerate the deterioration of buildings and objects. Labadi (2022) and others have argued that heritage practice must be integrated with #sustainable_development and climate adaptation. For diplomacy, climate risk creates both cooperation opportunities and tensions. States facing shared risks, such as those bordering the Mediterranean, have reasons to cooperate on climate adaptation for heritage sites. Small island states use their vulnerable heritage as evidence in climate negotiations. At the same time, decisions about which sites to protect, when funds are limited, will be politically charged. Not everything can be saved. Those choices will need to be made through processes that respect the communities most affected. 9.3 Armed Conflict The destruction of heritage during armed conflict has re-emerged as a defining issue. The Bamiyan Buddhas in Afghanistan, Palmyra in Syria, the Old City of Mosul in Iraq, Timbuktu in Mali, and heritage sites across Ukraine have all been damaged or destroyed in recent decades. In some cases, destruction was collateral. In others, it was deliberate and strategic. Cultural cleansing is a recognized dimension of contemporary conflict, aimed at erasing the presence of particular communities. The international response has combined legal action, emergency preservation, and documentation. The prosecution of Ahmad al-Mahdi at the International Criminal Court for the destruction of shrines in Timbuktu established that the deliberate destruction of cultural heritage can constitute a war crime. Rescue operations by UNESCO, ICOMOS, and coalitions of national institutions have salvaged what can be salvaged. Documentation efforts, often crowdsourced and digital, aim to preserve at least the memory of what has been lost. For strategic purposes, armed conflict is where the ethical stakes of heritage work are highest. It is also where the difference between principle and practice becomes clearest. States that condemn cultural destruction abroad while damaging heritage at home lose credibility. States that maintain principled positions and contribute practical help gain it. 9.4 The Coming Decades Looking ahead, three trends seem likely to intensify. First, competition among major powers will increase the strategic weight given to cultural heritage. Second, technological change will continue to expand what is possible in digitization and virtual engagement while also expanding what is possible in manipulation and forgery. Third, climate and conflict pressures will make the emergency dimension of heritage work more prominent. Students entering this field will need to combine technical competence with strong ethical commitments and a clear-eyed understanding of the political stakes. 10. Conclusion This article has argued that #cultural_heritage is now a central resource in international relations. States use it to build reputation, to construct partnerships, to project civilizational narratives, and to advance concrete foreign policy goals. The instruments are diverse, from bilateral preservation agreements to museum partnerships, from archaeological missions to digital platforms. The multilateral architecture, anchored by UNESCO, remains indispensable despite its limits. The politics of restitution and ownership are reshaping the field. Digital transformation, climate change, and armed conflict are producing new challenges that will define practice for decades. Three broader lessons stand out. First, #soft_power built on heritage is durable but demanding. It requires long-term investment, credible practice, and honest engagement with difficult histories. Shortcuts, exaggerations, and one-sided narratives eventually undermine the credibility they seek to build. Second, heritage diplomacy works best when it is genuinely reciprocal. Programs that treat partner countries as passive recipients produce weaker outcomes than those that build capacity, share resources, and support local leadership. Third, ethical seriousness is a strategic asset. States that can face contested pasts, return unjustly held objects, and support communities under threat gain influence that money alone cannot buy. For students, the field of #heritage_diplomacy offers unusual professional opportunities. It combines rigorous technical work with meaningful engagement in world affairs. It rewards those who can move between the specific detail of a conservation project and the broad frame of a bilateral relationship. It is also a field in which small actors, including small countries, small institutions, and individual scholars, can have disproportionate influence, because credibility and expertise carry weight beyond national scale. The past is not a fixed inheritance. It is a resource that each generation renegotiates. In the coming decades, the question is not whether heritage will matter in international relations. It will. The question is what kind of heritage practice we will build. If it is honest, participatory, and genuinely shared, it will help states build the kinds of partnerships that stabilize a fractious world. If it is deployed cynically, it will become one more source of grievance. The choice is not made once. It is made every time a monument is restored, an object returned, an exhibition mounted, a nomination submitted, and a partnership signed. That is a heavy responsibility. It is also a remarkable opportunity. References Blake, J. (2022). International Cultural Heritage Law. Second edition. Oxford: Oxford University Press. Clarke, D. (2020). Cultural diplomacy. In R. Marlin-Bennett (Ed.), Oxford Research Encyclopedia of International Studies. Oxford: Oxford University Press. https://doi.org/10.1093/acrefore/9780190846626.013.543 Grincheva, N. (2020). Museum Diplomacy in the Digital Age. London and New York: Routledge. https://doi.org/10.4324/9780429316555 Harrison, R., DeSilvey, C., Holtorf, C., Macdonald, S., Bartolini, N., Breithoff, E., Fredheim, H., Lyons, A., May, S., Morgan, J., and Penrose, S. (2020). Heritage Futures: Comparative Approaches to Natural and Cultural Heritage Practices. London: UCL Press. https://doi.org/10.14324/111.9781787356009 Hoelscher, S. and Alderman, D. H. (2021). Heritage and memory in a changing world. In C. Palmer and J. Tivers (Eds.), Heritage Tourism: A Contemporary Perspective. London: Routledge. Labadi, S. (2022). Rethinking Heritage for Sustainable Development. London: UCL Press. Labadi, S., Giliberto, F., Rosetti, I., Shetabi, L., and Yildirim, E. (2021). Heritage and the Sustainable Development Goals: Policy Guidance for Heritage and Development Actors. Paris: ICOMOS. Meskell, L. (2020). Toward a global comparative ethnography for UNESCO World Heritage. Journal of Field Archaeology, 45(sup1), S28 to S35. https://doi.org/10.1080/00934690.2020.1713284 Nisbett, M. (2020). The rise of cultural diplomacy in emerging powers. International Journal of Cultural Policy, 26(6), 743 to 758. Nye, J. S. (2021). Soft power: the evolution of a concept. Journal of Political Power, 14(1), 196 to 208. https://doi.org/10.1080/2158379X.2021.1879572 Sarr, F. and Savoy, B. (2021). Africa's Struggle for Its Art: History of a Postcolonial Defeat. Princeton: Princeton University Press. Original French edition published earlier and cited in later English-language reception. Silverman, H., Waterton, E., and Watson, S. (Eds.) (2022). Heritage in Action: Making the Past in the Present. Second edition. Cham: Springer. Smith, L. (2021). Emotional Heritage: Visitor Engagement at Museums and Heritage Sites. London and New York: Routledge. https://doi.org/10.4324/9781315099811 Vlassis, A. (2021). Global online platforms, COVID-19, and culture: The global pandemic, an accelerator towards which direction? Media, Culture and Society, 43(5), 957 to 969. https://doi.org/10.1177/0163443721994537 Winter, T. (2021). Geocultural power and the shifting logic of civilization. International Affairs, 97(5), 1547 to 1563. Winter, T. (2022). Geocultural Power: China's Quest to Revive the Silk Roads for the Twenty-First Century. Chicago: University of Chicago Press. #Cultural_Heritage #Soft_Power #Cultural_Diplomacy #Historical_Preservation #International_Relations #UNESCO #Heritage_Diplomacy #World_Heritage #Museum_Diplomacy #Restitution #Geocultural_Power #Silk_Roads #Digital_Heritage #Intangible_Heritage #Sustainable_Heritage #Global_Partnerships #Heritage_Policy #Heritage_Studies #Nation_Branding #Cultural_Policy #Public_Diplomacy #Colonial_Legacy #Museum_Studies #Heritage_Conservation #Global_Strategy