Educational Strategies for Clinical Supervision of Artificial Intelligence Use

Abdulnour, R. E., Gin, B., & Boscardin, C. K. (2025).

New England Journal of Medicine, 393(8), 786–797.

https://doi.org/10.1056/nejmra2503232

Human–computer interactions have been occurring for decades, but recent technological developments in medical artificial intelligence (AI) have resulted in more effective and potentially more dangerous interactions. Although the hype around AI resonates with previous technological revolutions, such as the development of the Internet and the electronic health record, the appearance of large language models (LLMs) seems different. LLMs can simulate knowledge generation and clinical reasoning with humanlike fluency, which gives them the appearance of agency and independent information processing. Therefore, AI has the capacity to fundamentally alter medical learning and practice. As in other professions, the use of AI in medical training could result in professionals who are highly efficient yet less capable of independent problem solving and critical evaluation than their pre-AI counterparts.

Such a challenge presents educational opportunities and risks. AI can enhance simulation-based learning, knowledge recall, and just-in-time feedback and can be used for cognitive off-loading of rote tasks. With cognitive off-loading, learners rely on AI to reduce the load on their working memory, a strategy that facilitates mental engagement with more-demanding tasks. However, off-loading of complex tasks, such as clinical reasoning and decision making, can potentially lead to automation bias (overreliance on automated systems and risk of error), “deskilling” (loss of previously acquired skills), “never-skilling” (failure to develop essential competencies), and “mis-skilling” (reinforcement of incorrect behavior due to AI errors or bias). These risks are especially troubling because LLMs operate as unpredictable black boxes; they generate probabilistic responses with low reasoning transparency, which limits assessment of their reliability. For example, in one study, more than a third of advanced medical students missed erroneous LLM answers to clinical scenarios.

The article is paywalled, unfortunately.

Here is my brief summary: 

This article addresses the challenge of supervising medical trainees who use artificial intelligence (AI) tools, particularly large language models (LLMs), in clinical reasoning. It notes that while AI can enhance learning through simulation, knowledge recall, and cognitive off-loading of rote tasks, off-loading complex tasks like clinical reasoning risks automation bias, "deskilling," "never-skilling," and "mis-skilling." LLMs are unpredictable black boxes with low reasoning transparency, making their reliability hard to assess.

The authors argue that critical thinking is foundational to adaptive practice in the age of AI. They propose the DEFT-AI framework (Diagnosis, Evidence, Feedback, Teaching, and recommendation for AI engagement) as a structured approach for educators to promote critical thinking during learner-AI interactions. The framework guides educators to probe the learner's clinical reasoning and AI use, evaluate supporting and opposing evidence, foster self-reflection, provide targeted teaching, and recommend safe AI engagement.

The article also describes two human-AI collaboration behaviors: centaur (strategic division of tasks, with human judgment leading) and cyborg (tight intertwining of user and AI throughout a task). Adaptive AI practice requires shifting between these modes based on task complexity and risk. The authors emphasize promoting AI literacy through evidence-based evaluation of AI tools and outputs, effective prompt engineering, and a "verify and trust" paradigm, concluding that AI interactions are here to stay and must be met with critical thinking and structured educational strategies.

Artificial Intelligence and Psychotherapy: Opportunities, Challenges, and Recommendations

Cooper, S., Ardlan, F., Gavazzi, J., et al. (2026).

A Report of the AI and Psychotherapy Work Group, 

Society for the Advancement of Psychotherapy, 

American Psychological Association, Division 29.

Executive Summary

Psychologists are already using artificial intelligence (AI) to draft progress notes, screen literature, generate case conceptualizations, and rehearse clinical skills with simulated patients. Technology has shaped professional psychology for decades, but no earlier development moved this quickly into so many parts of the work at once. A model can be given every word a patient has spoken and still have nothing at stake in what happens next. That disparity organizes the analysis in this report.

Emerging from the October 2025 SAP Board discussion of mega-level issues and opportunities for the Society for the Advancement of Psychotherapy, the Presidential Work Group on AI and Psychotherapy was established in 2026 by Society President Joshua Swift to examine AI’s implications for psychotherapy practice, supervision, education and training, and research. What follows is guidance from a Society work group. It is not APA policy and not a practice guideline.

The authors of the four domain sections worked independently and drew on different literatures. Their analyses converged: AI can substantially augment psychological work, but it cannot replace the human judgment, relationships, and accountability on which psychotherapy depends.

Six principles follow.

1. AI augments rather than replaces humans.
2. Human accountability remains essential.
3. AI output requires critical evaluation.
4. Implementation must be ethical and safe.
5. Human relationships remain central.
6. Continuous learning and governance are necessary.

Across all four domains the Work Group identified the same risks: automation bias, professional deskilling, cultural and demographic bias, threats to privacy and confidentiality, and overreliance on AI-generated recommendations. Beneath them sits a single concern. AI generates plausible interpretations from patterns in data, while psychotherapy requires understanding a particular person within a particular cultural, historical, and relational context. Pattern recognition can look like clinical understanding and be something else entirely.

The Work Group recommends evaluating any AI application against three questions: whether it enhances or erodes the human relationships the work depends on, whether it preserves psychologist accountability, and whether it respects the patient’s narrative integrity and cultural context. Evaluate a tool before it becomes embedded in practice, because a tool is far easier to decline than to remove.

Four practices follow. Psychologists should reason independently before consulting AI models whenever feasible, verify what AI produces, protect confidentiality in data handling, and review AI-assisted documentation before it enters a record.

Underlying these recommendations is a commitment to human dignity. Persons must not be reduced to data, classifications, or algorithmic predictions. The tools named in this report will be superseded and the studies cited will be replaced by better ones, but the question they raise will not date: what parts of psychological work can be augmented by technology, and what responsibilities cannot be delegated at all.

The entire work group report is here.

Would You Talk to Your Therapist’s AI ‘Twin’?

Isabel Woodford

Bloomberg.com

Originally posted September 11, 2026

Vanessa Marin is impressively responsive. The renowned sex therapist and New York Times bestselling author replies immediately to messages — even in the early morning hours. She doesn’t hesitate to reply when asked what issues she encounters most often in her work. While everyone’s sex life is different, she explains, the dilemmas people bring to her tend to be remarkably familiar.

Mostly it’s “things like whether it’s needy to want to feel desired by your partner,” she writes. “But it’s totally reasonable. You want your partner to want you — that’s not clingy, that’s human.”

This isn’t the real Vanessa Marin. It’s a digital twin she created using Delphi AI, a startup backed by Sequoia Capital that specializes in turning experts into chatbots. The bots, which are trained on material supplied by their human counterparts, including books, websites, videos and social media, have proved popular with celebrities including Arnold Schwarzenegger, who’s created a digital replica of his mind.

The article is here.

Ethical Dilemmas in End-of-Life Care: A Psychological and Anesthetic Perspective

Nmezi J. (2026).

Cureus, 18(6), e111202.

https://doi.org/10.7759/cureus.111202

Abstract

End-of-life care presents some of the most complex and emotionally challenging ethical dilemmas in modern medicine. For anesthetists and clinical psychologists, these dilemmas intersect at the critical juncture where physiological support meets psychological suffering. This editorial advances the novel argument that the psychological distress inherent to the dying process constitutes a specific clinical entity (existential suffering), which requires interdisciplinary management distinct from standard symptom control. It explores the major ethical challenges encountered in end-of-life care, including decisions about withholding and withdrawing life-sustaining treatment, do-not-resuscitate orders, medically administered nutrition and hydration, palliative sedation, and requests for assisted dying. The specific, practical roles of the anesthetist and clinical psychologist are detailed within a collaborative model, moving beyond generalities of "symptom management" and "grief support." Principles of biomedical ethics (autonomy, beneficence, non-maleficence, and justice) provide a framework for analyzing these dilemmas. Culturally specific decision-making frameworks for Nigerian and African contexts, including family-centric consensus models and communication scripts, are introduced. Practical recommendations are offered in the form of specific communication tools and assessment prompts for clinicians navigating these challenging situations and self-care to prevent moral distress and burnout.

The full editorial is here.

Here are some thoughts:

This editorial argues that existential suffering in end-of-life care is a distinct clinical entity requiring interdisciplinary management beyond standard symptom control. It outlines specific roles for anesthetists (e.g., translating physiological decline, managing terminal weaning) and clinical psychologists (e.g., dignity therapy, screening for demoralization) within a collaborative tiered model. Using biomedical ethics principles (autonomy, beneficence, non-maleficence, justice), it analyzes dilemmas like withholding treatment, DNR orders, palliative sedation, and assisted dying. The piece introduces culturally specific frameworks for Nigerian/African contexts, including family-centric consensus models and communication scripts. Practical tools, communication scripts, and self-care strategies are offered to help clinicians navigate these challenges and prevent moral distress.

Navigating increasing complexity in mental health practice: ETHICA-4P, a framework and toolkit to promote reflective skills for ethical clinical decision making

Calia, C., Grant, L., Guerra, C., & Reid, C. (2025).

Reflective Practice, 1–23.

https://doi.org/10.1080/14623943.2025.2586133

Abstract

Clinical practice for mental healthcare can be challenging. Often clinicians must make impactful decisions with little opportunity for reflection or consultation. In response to rich conversations with more than 900 local and global researchers and clinical colleagues (from more than 40 countries) about the increasing complexity of global research and clinical mental health practice, we developed the ETHICA-4P framework and toolkit for supporting clinicians in scaffolded reflection for ethical decision-making in daily practice (https://www.ethical-action.ed.ac.uk/clinical-practice). Ethical clinical interventions sit at the heart of the toolkit but practitioner-participants emphasised that ethical issues also arise during other stages of the clinical journey such as referral processes, assessment and during professional collaboration – with legacy effects long after therapy is complete. To navigate these multilayered challenges, we propose the ETHICA-4P framework and clinical toolkit centred on reflective analysis of ‘4P’s’:

• Place: Social, political, cultural, and historical factors shaping practice.

• People: Stakeholders involved in the ethical conflict, from clients to practitioners.

• Principles: Ethical guidelines and legal frameworks.

• Precedents: How past conflicts were addressed.

This paper reports on the practice-driven, co-design and iterative feedback process that led to the framework and toolkit, as well as spotlighting examples of clinical application.

Here are some thoughts:

U.S. psychologists operate in a high-stakes, high-diversity, high-burnout environment where ethical challenges regularly outpace formal guidance. ETHICA-4P offers a reflective, culturally attuned, practical supplement to the APA Ethics Code—one that supports both client welfare and clinician sustainability, and that can be used immediately in everyday practice, supervision, and training.

AI-enabled therapy platforms: Clinical promise, ethical challenges, and cultural considerations in mental health care

Srivastava, A., & Bita, M. (2026).

Asian Journal of Psychiatry, 123, 105087.

https://doi.org/10.1016/j.ajp.2026.105087

Abstract

The integration of artificial intelligence (AI) into mental health care presents significant opportunities to address gaps in access, affordability, and scalability, particularly in resource-constrained settings such as India. AI-enabled therapy platforms, including conversational agents and virtual therapists, promise personalized and continuous psychological support. However, alongside these clinical advantages, critical ethical and contextual challenges emerge, including concerns related to diagnostic validity, data privacy, informed consent, therapeutic alliance, and cultural misalignment.

This narrative review critically synthesizes contemporary global and regional literature (2019–2025) to examine the clinical potential, ethical risks, and cultural implications of AI-driven psychotherapy. Drawing on a psychologist-centered perspective, the review emphasizes the necessity of rigorous validation, transparent governance, culturally adaptive design, and a collaborative therapist–AI model rather than full automation. Particular attention is given to the Indian sociocultural context, where mental health experiences are deeply embedded in relational, linguistic, and indigenous psychological frameworks.

The review argues that ethically grounded and culturally informed AI-enabled therapy platforms can enhance mental health service delivery when integrated as adjunctive tools within professional care systems. By situating AI innovation within India’s pluralistic psychological landscape, this paper offers a context-sensitive, evidence-informed set of recommendations for responsible and clinically meaningful AI integration in mental health care.

Highlights

• Reviews the clinical applications of AI-enabled therapy platforms within mental health care.

• Examines the ethical risks and governance challenges associated with AI-supported psychotherapy.

• Emphasizes a psychologist-centered model of AI use that incorporates a human-in-the-loop approach.

• Discusses cultural considerations relevant to Asian and Indian contexts.

• Proposes a balanced framework that integrates AI with the therapeutic alliance.

Racial, Ethnic, and Cultural Expressions of Interpersonal Psychological Theory of Suicide (RECEIPTS): An Integrated Model of Structural Racism and Suicide RiskIn Section American Psychologist

Adams, L. B.,  et al. (2025).

American Psychologist, 81(5), 597–610.

https://doi.org/10.1037/amp0001545

Abstract

Suicide risk is a significant public health concern for individuals and communities across the United States, and the rates of suicidality are disproportionately rising for Black Americans. Recent frameworks have articulated the significance of structural racism as a mechanism that may explain the increasing rates of suicide among Black Americans, in part, through its exacerbating effects on salient risk conferring pathways. However, existing scholarship in this area has been developmentally limited in scope and does not specify how structural racism operates as a macrolevel determinant of suicide across the lifespan. To address this gap, we present the Racial, Ethnic, and Cultural Expressions of Interpersonal Psychological Theory of Suicide (RECEIPTS), which highlights how structural racism catalyzes suicide risk for Black Americans. The RECEIPTS model supports and extends tenets of the interpersonal theory of suicide and provides a generalizable and comprehensive framework to understand the complex and intersecting factors that contribute to suicidality among Black Americans across the life course. The RECEIPTS framework highlights structural racism’s impact on suicide risk, offering implications for culturally informed prevention, research, and clinical practice.

Public Significance Statement

The article addresses the interplay between structural racism and suicide risk among Black Americans by introducing the Racial, Ethnic, and Cultural Expressions of Interpersonal Psychological Theory of Suicide. This work highlights and presents an explanatory framework to address a critical gap in the understanding of suicide prevention by emphasizing structural racism as a central factor in delineating mental health disparities among Black Americans.

The basic psychological needs in excellencism and perfectionism: A dual perspective with the need-as-motives and the need-as-nutriments frameworks

Andrade, G., et al. (2025).

Personality and Individual Differences, 

245, 113286.

https://doi.org/10.1016/j.paid.2025.113286

Abstract

Perfectionism has been theorized as a risk factor for psychological need frustration. However, past studies on basic psychological needs often reported ambiguous and unexpected findings for perfectionistic standards. The Model of Excellencism and Perfectionism (MEP) recently distinguished between perfectionistic standards and the pursuit of high yet attainable standards (excellencism). This study investigated their distinct associations with basic psychological needs, using measures taken from the need-as-motives and the need-as-nutriments perspectives. Young adults (n = 305) completed the Scale of Perfectionism and Excellencism and various measures of need-related constructs. A multivariate multiple regression supported the hypothesis that excellencism and perfectionism are differentially linked with psychological needs. Excellencism was positively associated with three approach-oriented motives (need for achievement, affiliation, power) and satisfaction with the need for autonomy, relatedness, and competence. Conversely, pursuing perfectionistic standards was positively linked to two avoidance-oriented motives (e.g., fear of failure and losing control) and frustration with the three basic psychological needs. These findings reconcile research and theories by showing that pursuing perfection is not associated with adaptive psychological needs. Perfectionistic standards are linked to two avoidance-oriented motives (i.e., fear of failure and losing control) and frustration of basic psychological needs when properly distinguished from excellencism.

A pdf is here.

Here are some thoughts:

Clinically, these findings suggest that interventions should promote excellencism (the pursuit of high yet attainable standards) rather than attempting to harness perfectionism as a “healthy” or adaptive trait. Perfectionistic standards offer no additional motivational benefits over excellencism (e.g., no greater need for achievement or need satisfaction) while reliably adding need frustration (competence, relatedness, and autonomy) and avoidance-based fears (failure, losing control). This reframes perfectionism as an unnecessary risk factor rather than a useful drive. Practitioners should also note perfectionists’ antagonistic motive profile, specifically heightened need for power paired with fear of losing control, which may manifest as hypercompetitiveness and interpersonal hostility. Finally, clinicians should reinterpret past research cautiously, as many previously “adaptive” effects attributed to perfectionistic standards were likely artifacts of conflating perfectionism with excellencism.

Prompt Engineering in Clinical Practice: Tutorial for Clinicians

Liu, J., Liu, F., Wang, C., & Liu, S. (2025).

Journal of Medical Internet Research, 27, e72644.

https://doi.org/10.2196/72644

Abstract

Large language models (LLMs), such as OpenAI’s GPT series and Google’s PaLM, are transforming health care by improving clinical decision-making, enhancing patient communication, and simplifying administrative tasks. However, their performance relies heavily on prompt design, as small changes in wording or structure can greatly impact output quality. This presents challenges for clinicians who are not experts in natural language processing (NLP). This tutorial combines prompt engineering techniques tailored for clinical use, covering methods like zero-shot prompting, one-shot prompting, few-shot prompting, chain-of-thought prompting, self-consistency prompting, generated knowledge prompting, and meta-prompting. We provide actionable guidance on defining objectives, applying core principles, iterative prompt refinement, and integration into interoperable electronic health record (EHR) systems. This framework helps clinicians leverage LLMs to improve decision-making, streamline documentation, and enhance patient communication while maintaining ethical standards and ensuring patient safety.

This is an excellent start your journey on prompt engineering and prompt competency.

Auditing the AI Auditors: A Framework for Evaluating Fairness and Bias in High Stakes AI Predictive Models

Landers, R. N., & Behrend, T. S. (2023).

American Psychologist, 78(1), 36–49.

https://doi.org/10.1037/amp0000972

Abstract

Researchers, governments, ethics watchdogs, and the public are increasingly voicing concerns about unfairness and bias in artificial intelligence (AI)-based decision tools. Psychology’s more-than-a-century of research on the measurement of psychological traits and the prediction of human behavior can benefit such conversations, yet psychological researchers often find themselves excluded due to mismatches in terminology, values, and goals across disciplines. In the present paper, we begin to build a shared interdisciplinary understanding of AI fairness and bias by first presenting three major lenses, which vary in focus and prototypicality by discipline, from which to consider relevant issues: (a) individual attitudes, (b) legality, ethicality, and morality, and (c) embedded meanings within technical domains. Using these lenses, we next present psychological audits as a standardized approach for evaluating the fairness and bias of AI systems that make predictions about humans across disciplinary perspectives. We present 12 crucial components to audits across three categories: (a) components related to AI models in terms of their source data, design, development, features, processes, and outputs, (b) components related to how information about models and their applications are presented, discussed, and understood from the perspectives of those employing the algorithm, those affected by decisions made using its predictions, and third-party observers, and (c) meta-components that must be considered across all other auditing components, including cultural context, respect for persons, and the integrity of individual research designs used to support all model developer claims.

Public Significance Statement

Although artificial intelligence (AI) is now being used to make decisions about people’s employment, education, healthcare, and experiences with law enforcement, external evaluators do not often agree on what is necessary to show that an AI is “unbiased” or “fair.” This is in part because “bias” and “fairness” mean different things to different people. We created a framework for auditing that respects these differences in pursuit of better, fairer AI.

Effect of psychotropic medications on suicide-related outcomes: a systematic review and meta-analysis of observational studies

Kozhevnikova, S.,et al. (2026).

EClinicalMedicine, 93, 103800.

https://doi.org/10.1016/j.eclinm.2026.103800

Abstract

Background: Psychiatric disorders are associated with increased risk of suicide-related outcomes, and the impact of pharmacological treatments on these outcomes is uncertain. Although randomised controlled trials are the main approach to evaluate efficacy, they may not provide externally valid results for suicide prevention in psychiatric populations. Thus, we aimed to synthesise the evidence on the effect of psychotropic medications on suicide-related outcomes from observational studies.

Methods: In this systematic review and meta-analysis, we systematically searched Ovid (MEDLINE, Embase, APA PsychArticles, AMED, BIOSIS, Global Health, PsycINFO), and Web of Science Core Collection from database inception to 8 December 2025 for pharmacoepidemiological and other observational studies on suicide-related outcomes in people treated with the main types of psychotropic medications: antidepressants, antipsychotics, mood stabilisers (including antiepileptics), and medications for anxiety (anxiolytics), attention deficit and hyperactivity disorder (ADHD), and substance use disorder (SUD). We included primary studies involving adults with common psychiatric diagnoses (schizophrenia spectrum disorders, bipolar disorder, depressive disorders, and personality disorders), who were prescribed medication and a comparison sample with the same diagnosis without prescribed medication (between-individual studies) or the same individuals during a non-prescription period (within-individual studies). We excluded studies that did not report psychiatric diagnoses and from selected samples. Outcomes were suicide attempts/self-harm and suicide mortality. We pooled effect sizes as odds ratios (OR), hazard ratios (HR) or risk ratios (RR) using random-effects models and assessed study quality using NOS and QUIPS tools. Study protocol was registered with PROSPERO, CRD42024515794.

Findings: Of 5653 records identified, 48 independent studies from 13 countries based on more than 6 million people (47% male) met inclusion criteria. Across the main diagnostic categories and 70 individual medications examined, associations with reducing risk of suicide mortality were found for second generation antipsychotics in schizophrenia spectrum disorders: clozapine (OR = 0.40; 0.36-0.60; I2 = 60%, moderate certainty), olanzapine (OR = 0.53; 0.39-0.71; I2 = 34%, high certainty), quetiapine (OR = 0.75; 0.58-0.96; I2 = 0%, high certainty), and zuclopenthixol (OR = 0.44; 0.30-0.63; I2 = 0%, high certainty). In schizophrenia, second generation antipsychotics were also associated with reduced risks of suicide attempts: olanzapine (OR = 0.76; 0.60-0.98; I2 = 84%, moderate certainty) and risperidone (OR = 0.61; 0.52-0.72; I2 = 57%, moderate certainty). In bipolar disorder, lithium (OR = 0.38; 0.28-0.50; I2 = 67%, moderate certainty) and valproic acid (OR = 0.66; 0.59-0.75; I2 = 0%, high certainty) were associated with lower suicide risks, and lithium was also associated with lower risks of suicide attempts (OR = 0.60; 0.44-0.82; I2 = 92%, moderate certainty). In depression, associations with lower risk of suicide mortality for selective serotonin reuptake inhibitors (SSRIs) (OR = 0.61; 0.47-0.81; I2 = 23%, high certainty) and tricyclic antidepressants (OR = 0.68; 0.59-0.78; I2 = 0%, high certainty) were found. Benzodiazepines were associated with higher risk of suicide mortality in most diagnostic categories, except depression. There was some evidence for publication bias for lithium in bipolar disorder and clozapine in schizophrenia spectrum disorders, leading to more papers reporting lower risks of suicide-related outcomes. The risk of bias in included studies was low in 47 studies, moderate in one study, and certainty of evidence was moderate.

Interpretation: There is evidence of varying effects of psychotropic medication on the risk of suicide-related outcomes across different psychiatric disorders. The appropriate use of prescribed medications in people with high risks of suicide-related outcomes is an important suicide prevention strategy. Findings are not causal, and limitations include the observational nature of included studies, risk of residual confounding, high heterogeneity for some outcomes, and moderate quality of the evidence.

My summary:

This systematic review and meta-analysis of observational studies demonstrates that the impact of psychotropic medications on suicide-related outcomes varies significantly across different psychiatric diagnoses. 

In schizophrenia spectrum disorders, second-generation antipsychotics—specifically clozapine, olanzapine, quetiapine, and risperidone—are associated with a reduced risk of suicide mortality and/or suicide attempts. 

For individuals with bipolar disorder, both lithium and valproic acid show strong associations with lowered suicide-related risks.

In depressive disorders, selective serotonin reuptake inhibitors (SSRIs) and tricyclic antidepressants are associated with a decreased risk of suicide mortality. 

Conversely, benzodiazepines are consistently linked to an increased risk of suicide-related outcomes across most diagnostic categories.

A pdf can be downloaded here.

The use of CAMS and DBT to effectively treat patients who are suicidal

Jobes, D. A., & Rizvi, S. L. (2024).

Frontiers in psychiatry, 15, 1354430.

https://doi.org/10.3389/fpsyt.2024.1354430

Abstract

Around the world, suicide ideation, attempts, and deaths pose a major public and mental health challenge for patients (and their loved ones). Accordingly, there is a clear need for effective clinical treatments that reliably reduce suicidal thoughts and behaviors. In this article, we review the Collaborative Assessment and Management of Suicidality (CAMS) and Dialectical Behavior Therapy (DBT), two clinical treatments that rise to the highest levels of empirical rigor. Both CAMS and DBT are now supported by randomized controlled trials (RCTs), with independent replications, and meta-analyses. There are also supportive data related to training clinical providers to use CAMS and DBT with adherence. RCTs that investigate the use of both interventions within clinical trial research designs and the increasing use of these complementary approaches within routine clinical practice are discussed. Future directions for research and clinical use of CAMS and DBT are explored as means to effectively treat suicidal risk.

Here are some thoughts:

This review's real value is putting CAMS and DBT side by side on equal empirical footing, showing them as complementary rather than competing: CAMS for acute, driver-focused ideation in 4-12 sessions, DBT for chronic, multi-attempt presentations tied to emotion dysregulation over 6+ months. The evidence patterns mirror this division nicely. The Swift et al. meta-analysis found CAMS significantly reduces suicidal ideation and hopelessness but had no significant effect on attempts specifically, while DeCou et al.'s DBT meta-analysis found the reverse, reduced self-directed violence but no significant ideation effect. The Pistorello SMART trial adds a useful clinical heuristic: patients with no attempt history did better with CAMS, while those with multiple attempts and borderline features responded better to treatment as usual than to CAMS, consistent with DBT's stronger footing for that chronic population.

A couple of things worth flagging. Several CAMS trials were genuinely mixed, underpowered, or favored treatment as usual early in follow-up, and Jobes discloses a financial interest as founder of CAMS-care, LLC, which doesn't invalidate the independently replicated RCTs but is worth naming given how the mixed findings are framed. The Hope Institute section is the most practically interesting piece for service delivery: a next-day-appointment outpatient model achieving stabilization in 5-6 weeks, positioned against the paper's opening point that inpatient hospitalization itself lacks strong evidence outside the immediate post-attempt window.

Ethical Decision-Making Guidelines for Mental Health Clinicians in the Artificial Intelligence (AI) Era

Pillay Y. (2025).

Healthcare (Basel, Switzerland), 13(23), 3057.

https://doi.org/10.3390/healthcare13233057

Abstract

The meteoric rise in generative AI has created both opportunities and ethical challenges for the mental health disciplines, namely in clinical mental health counseling, psychology, psychiatry, and social work. While these disciplines have been grounded in well-established ethical principles such as autonomy, beneficence, justice, fidelity, and confidentiality, the exponential ubiquity of AI in society has rendered mental health professionals unsure as to how to navigate ethical decision making in the AI era. The author proposes a preliminary ethical framework which synthesizes the code of ethics of the American Counseling Association (ACA), the American Psychological Association (APA), the American Medical Association (AMA), and the National Association of Social Workers (NASW), which is then organized around five pillars: (i) autonomy and informed consent; (ii) beneficence and non-malfeasance; (iii) confidentiality, privacy, and transparency; (iv) justice, fairness and inclusiveness; and (v) fidelity, professional integrity, and accountability. These pillars are juxtaposed with AI ethical guidelines developed by multinational organizations, governmental and non-governmental entities, and technology corporations. The resulting integrated ethical framework provides a practical cogent structure that mental health professionals can use when navigating this uncharted terrain. A case study based on the proposed ethical framework and strategies that clinical mental professionals can consider prior to incorporating AI into their clinical repertoire are offered. Limitations of the framework and its implications for future research are addressed.

A pdf is here.

Here are some thoughts.

This article proposes an ethical decision-making framework for mental health clinicians integrating generative AI into practice, synthesizing codes from ACA, APA, AMA, and NASW with AI guidelines from global organizations. The framework organizes around five pillars: autonomy and informed consent, beneficence and non-malfeasance, confidentiality and transparency, justice and inclusiveness, and fidelity and accountability. A case study illustrates the framework's application with a refugee client using an AI chatbot as a between-session tool. The author also provides a practical checklist for clinicians before adopting AI, covering consent, vendor vetting, data security, competence, inclusivity, and transparency. Limitations include the framework's untested status, Western-centric bias, and rapid AI evolution, with recommendations for further empirical validation and curriculum updates.

Prejudiced Patients: Ethical Considerations for Addressing Patients’ Prejudicial Comments in Psychotherapy

Mbroh, H., et al. (2020).

Professional Psychology:

Research and Practice, 51(3), 284–290.

https://doi.org/10.1037/pro0000280

Abstract

Psychologists will often encounter patients who make prejudiced comments during psychotherapy. Some psychologists may argue that the obligations to social justice require them to address these comments. Others may argue that the obligation to promote the psychotherapeutic process requires them to ignore such comments. The authors present a decision-making strategy and an intervention based on principle-based ethics for thinking through such dilemmas.

Public Significance Statement

This article identifies ethical principles psychologists should consider when deciding whether to address their patients’ prejudicial comments in psychotherapy. It also provides an intervention strategy for addressing patients’ prejudicial comments.

A pdf is here.

Here are some thoughts:

This article examines the ethical dilemmas psychologists face when patients make prejudicial comments during psychotherapy, highlighting the tension between the obligation to promote social justice (generalized beneficence) and the duty to prioritize the patient’s therapeutic well-being (beneficence and nonmaleficence). Using a principle-based ethics framework, the authors analyze various clinical scenarios to demonstrate how practitioners can weigh conflicting ethical obligations, such as respecting patient autonomy while mitigating harm to marginalized groups or the therapeutic alliance. To navigate these complex situations, the article recommends proactive education, self-reflection, and professional consultation, ultimately providing a structured, non-confrontational intervention strategy—empathizing, highlighting the personal impact of the beliefs, and inviting collaborative exploration—to address prejudice in a way that minimizes therapeutic rupture and fosters patient self-understanding.

Therapist Positionality: A Cornerstone of Antiracist Clinical Practice

Ashley, W. (2025).

Open Journal of Social Sciences, 13(11), 58–66.

https://doi.org/10.4236/jss.2025.1311004

Abstract

Therapist positionality is a foundational principle of antiracist clinical practice, yet it remains underemphasized in practitioner education, training, and ongoing professional development. Positionality refers to the interplay of a therapist’s intersecting identities, including race, gender, class, sexuality, ability, and professional status, and how these shape power, perception, and relational dynamics. This article examines the role of positionality across clinical, collegial, and supervisory relationships, offering strategies to promote reflexivity, cultural humility, and relational accountability. Using positionality, clinicians can move beyond cultural competence into an ongoing, embodied practice of cultural humility, ethical engagement, and cultural responsiveness, particularly with clients and colleagues from marginalized communities.

Here are some thoughts:

Ashley argues that therapist positionality, defined as the critical examination of how a clinician’s intersecting identities, power, and privilege influence the therapeutic relationship, is essential to moving beyond traditional notions of therapist neutrality toward genuine antiracist and culturally responsive practice. Moving beyond superficial disclosures, the paper details how authentic positionality requires ongoing intrapsychic self-reflection and interpersonal accountability across clinical, collegial, and supervisory dynamics. By explicitly naming power differentials and replacing the myth of the objective practitioner with embodied cultural humility, clinicians can foster greater trust, psychological safety, and relational repair with clients and colleagues from marginalized communities.

Think First, AI Second: Why Human Judgment Still Matters in Psychology

In this episode of Beyond the Couch, Dr. Ernest Wayde sits down with Dr. John Gavazzi, board-certified clinical psychologist, ethics educator, and former President of the Pennsylvania Psychological Association, to explore how psychologists can responsibly integrate artificial intelligence into clinical practice without compromising professional judgment or quality of care. 

Drawing decades of experience in ethics, mental health law, and clinical decision-making, Dr. Gavazzi explains why AI should enhance, not replace, the thinking process of psychologists. He introduces the concept of "Think First, AI Second," arguing that clinicians must remain the originators of thought, using AI only after applying their own expertise and clinical reasoning. 

The conversation explores the growing use of AI by both clinicians and clients, examining why people are turning to AI for emotional support, how psychologists can adapt when clients bring ChatGPT into therapy, and what must be preserved as AI becomes increasingly embedded in mental health care. Dr. Gavazzi also discusses cognitive offloading, automation bias, ethical accountability, AI literacy, and practical ways clinicians can redesign their workflows while maintaining professional competence. 

Ultimately, this episode reinforces that while AI is becoming an invaluable tool for psychology, human connection, ethical responsibility, and clinical judgment remain irreplaceable. 

Takeaways 

-Psychologists Should Think First and Use AI Second. 

-AI Can Support Clinical Work, But It Cannot Replace Human Judgment. 

-Cognitive Offloading Can Weaken Clinical Skills Over Time. 

-AI's Fluency Should Never Be Mistaken for Accuracy. 

-Clients Are Already Bringing AI Into Therapy Sessions. 

Development and Efficacy of an Ethical Decision-Making Tool for Training Clinical Psychologists

Hunt, M. G., et al. (2024).

Training and Education in Professional Psychology,

18(4), 284–296. https://doi.org/10.1037/tep0000467

Abstract

We developed an ethical decision-making tool to assist in training clinical psychologists that had four goals: (a) to mirror the way psychologists naturally think about ethical dilemmas; (b) to facilitate thorough analysis by using a simple, guided checklist format; (c) to provide quick access to intuitively organized explanations and examples of each major domain of ethical challenges and principles; and (d) to facilitate good documentation and record-keeping of the decision-making process. Study 1 surveyed clinical psychologists and asked them to list the top 10–15 ethical considerations they encounter in practice, and then used those categories to help develop the tool. In Study 2, having access to two different versions of the tool significantly improved the quality of analysis of a standardized clinical vignette, particularly for individuals still in training. In Study 3, a streamlined version of the tool significantly improved the quality of graduate trainees’ analysis of the standardized clinical vignette and was faster to use. In Study 4, we demonstrated the efficacy of the tool in helping trainees analyze ethical dilemmas they had actually experienced. In Study 5, faculty at doctoral programs who teach graduate ethics courses provided dilemmas and then graded the responses of trainees who analyzed them either with or without the tool. Analyses completed with the tool received far better grades on average than those completed without the tool. This work is the first that we know of to develop an evidence-based, empirically supported tool for training future psychologists in ethical practice.

Impact Statement

This study describes the development of a novel ethical decision-making tool for use in training clinical psychology graduate students. The study shows that the tool helps trainees improve their analysis of complex ethical dilemmas that arise in clinical practice.

The EDM tool is here.

Practical Frameworks for Integrating Artificial Intelligence in Forensic Psychology

Rilen, S. (2026).

Journal of Forensic Psychology Research

and Practice, 1–39.

https://doi.org/10.1080/24732850.2026.2630383

The article is paywalled. Please contact the author for a copy of this article.

Abstract

The American Psychological Association (APA) provides broad guidance for generative Artificial Intelligence (gAI) use, yet forensic practice requires a specialized context for adversarial settings. This article proposes a four-step operational framework for ethical gAI integration grounded in forensic admissibility. It establishes operational mandates and readiness assessments followed by a risk-stratification model for tool selection distinguishing convergent from divergent tasks. It differentiates substantive from peripheral uses to guide disclosure decisions, and addresses admissibility considerations under Daubert and Frye. Sample disclosure language and visual aids help experts preserve credibility through methodological verification and systematic procedural rigor, rather than algorithmic trust.

Here are some thoughts:

The article establishes that while generative AI offers significant potential to enhance forensic psychology, its integration must be governed by a strict, four-step framework that prioritizes practitioner accountability, methodological transparency, and data security. Central to this is the principle of non-delegation of cognition, which mandates that forensic experts retain full ownership of all analytical reasoning and conclusions, using gAI only as an augmentative tool rather than a substitute for professional judgment. All gAI output must be treated with a zero-trust verification approach, requiring independent validation against primary sources to guard against hallucinations, omissions, and algorithmic bias. Furthermore, the framework draws a critical distinction between substantive use—which materially influences opinions and requires proactive disclosure in court reports—and peripheral use, which involves administrative support and only requires informed consent. Given the adversarial legal context, practitioners must demonstrate technical competence to explain gAI's probabilistic nature and limitations, ensuring their methodology can withstand scrutiny under Daubert or Frye standards. Data privacy is paramount, necessitating enterprise-grade tools with robust security pillars, including data quarantine and encryption, to preserve legal privilege. Ultimately, the article emphasizes that forensic practitioners are the gatekeepers of their methodology, and by embracing rigor, transparency, and continuous self-assessment, they can harness gAI's benefits while safeguarding the integrity of the justice system.

Privilege in the room: Training future psychologists to work with power, privilege, and intersectionality within the therapeutic relationship.

Wright, A. J., et al. (2025).

Psychotherapy, 62(1), 82–89.

https://doi.org/10.1037/pst0000563

Between the racial reckoning of 2020 and wider spread policy development that is explicitly homophobic and transphobic, there have been consistent and resurgent calls for clinicians to address aspects of power and privilege in psychotherapy. This is especially important in a field that continues to be largely White, cisgender, and heterosexual (not to mention abled, socioeconomically privileged, and privileged in many other aspects of human diversity). However, too few models for how to accomplish this in actual practice are offered in the literature. Further, while there is little guidance for clinicians on how to address power, privilege, and intersectionality in the therapy room, there is even less direction for how to train those learning to be clinicians to do this from the start. The purpose of this article is to translate existing knowledge into a framework for supervisors to guide trainees’ application in psychotherapy. The article provides an overview of social location, including an analytic framework, as well as a set of practical steps for supervisors and trainees.

Impact Statement

Question: This article addresses how to train future psychologists to address issues of power, privilege, and intersectionality in their clinical work utilizing a structural frame. Findings: In addition to foundational definitions and concepts, the article provides guidance for supervisors, trainees, and clinicians on how to address power, privilege, intersectionality, and structural dynamics in clinical and supervisory practice. 

Meaning: Being vigilant and deliberate in addressing power, privilege, and intersectionality in clinical practice is crucial for building successful therapeutic and supervisory relationships, so employing the strategies shared will ultimately enhance clinical outcomes and create deeper understanding of the impact of structural oppression on the therapy enterprise. Next Steps: Training programs and supervisors must be vigilant and held accountable for engaging in the challenging work of addressing power, privilege, and intersectionality, utilizing a structural frame, to enhance clinical outcomes. 

A pdf can be found here.

The SHAPE Model: A Metacognitive Framework for Ethical Decision-Making

Gavazzi, J. (2026).

The Pennsylvania Psychologist, (86)2, 32-35.

Clinical Impact Statement

What Problem Does This Address for Clinicians?

When facing urgent, complex ethical dilemmas—such as managing high-risk safety concerns, unexpected disclosures, or boundaries in high-stress settings—clinicians often default to quick intuition because traditional ethical decision-making models are too lengthy or rigid to use in the moment.

How Does This Help in Everyday Practice?

The SHAPE model (Scrutinize, Hypothesize, Analyze, Perform, Evaluate) gives clinicians a fast, reliable mental framework to slow down just enough to avoid blind spots. It builds directly on a therapist’s clinical instincts while adding crucial guardrails:

• Prevents Knee-Jerk Actions: Prompts clinicians to step back and evaluate multiple options before acting under stress.

• Catches Unintended Consequences: Uses explicit safety checks (Contraindication Rule) to identify potential harms to the client or therapeutic alliance before a plan is executed.

• Works Under Pressure: Adapts gracefully to time-sensitive clinical scenarios without compromising ethical standards.

Bottom Line for Patient Care:

Using SHAPE helps therapists translate abstract ethics codes into practical, defensible, and compassionate clinical actions. Over time, reflecting on past decisions (Evaluate) sharpens clinical judgment, protects client welfare, and strengthens the therapeutic alliance.

A pdf is here.

An AI Perspective on Counseling Supervision

Brinck, E. A., et al. (2026).

Behavioral Sciences, 16(6), Article 1038. 

https://doi.org/10.3390/bs16061038

Abstract

The increased use of technology-assisted distance counseling practices is one result of COVID’s impact on behavioral health, including in counselor education and the delivery of supervision. First, technology-assisted distance supervision needed for “real time” communication grew. Furthermore, there is an emergence of artificial intelligence (AI) technologies that have the potential to contribute to aspects of supervision; however, current evidence remains emerging, context-dependent, and at times mixed, warranting cautious interpretation of their effectiveness. The article offers an overview of using AI in clinical supervision, examines the benefits and potential concerns of AI from different perspectives, and considers the significance of using AI in counseling supervision. The role of AI is discussed as applied to counseling supervision including the use of AI tools, such as chatbots and reasoning AI, to detect and track sessions, note behavioral and emotional cues, aid/monitor communication and feedback, while also attending to ethical and legal consideration for its use. The article will report a range of benefits for supervisors and trainees using AI—for example, by enhancing data-driven supervision decisions, analyzing feedback trends, providing more efficient administrative monitoring, flexible/remote support, skill development, and promoting ethical decisions and self-reflection. Special attention is given to the challenges of using AI in supervision, including risks of undervaluing intuition and qualitative insights, potential for algorithms to reinforce systemic biases, risks of replacing human interaction, as well as non-compliance with HIPAA, FERPA, and ethical guidelines in data storage and privacy. The article will discuss privacy concerns, depersonalized feedback, and increased judgment-driven anxiety despite needed empathy when using AI as a tool for clinical supervision. Recommendations will also be offered for effective, ethical integration of AI in counseling supervision.

Here are some thoughts:

This is a comprehensive review, and its most useful feature for a clinical audience is how thoroughly it maps existing professional guidance onto the specific problem of AI-assisted supervision. The authors walk through the NBCC's 2024 clinical tenets and the ACA's 13 recommendations in detail, which gives the piece real practical value as a reference document, something a supervisor could actually consult when drafting an informed consent form or a supervision contract that addresses AI use. The case example of Lexi and Jalen, where AI-flagged session notes surfaced a pattern of the trainee talking over a client's frustration, is a nice illustration of AI functioning as intended: not replacing supervisory judgment but generating material for the supervisor to process relationally with the supervisee.

The empirical base here is thinner than the framework suggests, and the authors are reasonably candid about this.

The most clinically substantive contribution is the discussion of the supervisory working alliance (SWA) as the issue actually at risk. The authors argue that AI-mediated feedback, delivered without the moderating effect of a trusted relationship, risks landing as more punitive or deficit-focused than the same content delivered by a supervisor who knows the trainee's developmental stage and history. That's a sharper articulation than most AI-and-supervision pieces manage, and it connects naturally to the Integrated Developmental Model framework they invoke early on: a supervisee's readiness to receive AI-flagged feedback without a supervisor's relational buffering plausibly varies by developmental level in ways this piece gestures at but doesn't fully develop.

Where I'd flag some limitation: the literature search was explicitly broad and non-systematic ("AI in human services" across a handful of databases), which the authors acknowledge as a scope limitation given how new this area is. And several of the ethical and legal considerations, informed consent language, data ownership, HIPAA and FERPA compliance, are more thorough in their treatment of what should happen than in documenting what is currently happening in supervision practice. The piece reads as a strong, well-organized synthesis and practical guide rather than an evidence base, which is an entirely fair thing for a review in this stage of the literature to be, but worth naming for readers expecting empirical weight.

The Role of Artificial Intelligence in Clinical Psychology: How AI and NLP Systems Are Reshaping Psychological Interventions. A Systematic Review

Orrù, L., & Mannarini, S. (2026). 

Clinical Psychology & Psychotherapy, 

33(2), e70242. 

https://doi.org/10.1002/cpp.70242

Abstract

Artificial Intelligence (AI) technologies are rapidly evolving and their integration into psychological practices has progressively expanded, offering new tools for diagnosis, treatment and therapeutic monitoring. This review examines the transformative role of AI, particularly Natural Language Processing (NLP) systems, in reshaping clinical psychology and digital mental health interventions (DMHIs). In particular, it explores how AI and NLP can facilitate human-machine interaction in therapy by analysing how language is used within clinical conversations and providing personalized, real-time interventions. Following PRISMA guidelines, a systematic review of literature from 2019 to 2025 identified 17 studies that met inclusion criteria, emphasizing AI's use in psychological assessment and intervention. The review focuses on two key aspects: the functions and applications of NLP-based systems in clinical practice and the advantages and benefits they offer for both psychologists and patients. Findings suggest that NLP-driven AI systems enhance both patient engagement and clinician efficiency, offering scalable, cost-effective solutions that improve access and personalization. However, challenges remain, including ethical concerns around data privacy, lack of standardization, limited generalizability across disorders and reduced human empathy. Moreover, current systems are primarily designed for well-defined conditions like anxiety and depression, with limited applicability to complex or comorbid psychological presentations. This review underscores the importance of supervised, ethically governed AI implementation. While AI holds substantial promise in augmenting clinical psychology, its success depends on maintaining human oversight, ensuring transparency and establishing shared scientific and ethical standards across the psychological community.

Summary

• Through the systematic review of the existing literature regarding the use of AI in clinical psychology, practitioners could know how AI and NLP systems are (re)shaping psychological interventions to improve them for digital mental health (DMH).
• The study describes characteristics, features and tasks of AI systems in clinical psychology, the main techniques currently adopted, from which practitioners can find information to choose the best tools for their work objectives, adapting to specific individual psychological needs.
• Practitioners could evaluate AI systems' impact on DMH promotion: beneficial effects of AI systems for patients' mental health and related challenges. Indeed, while on one hand the implementation of AI in clinical psychology has demonstrated significant improvements in psychological well-being, on the other hand some studies have reported less conclusive findings.

The unintended negative consequences of artificial intelligence use for psychologists

van Zyl, L. E. (2026).

Frontiers in Psychology, 17, Article 1729050.

https://doi.org/10.3389/fpsyg.2026.1729050

Abstract

The rapid integration of artificial intelligence (AI) into psychological practice is transforming how psychologists reason, document, and deliver care. While benefits such as improved diagnostic accuracy and administrative efficiency are widely recognized, the unintended consequences for psychologists themselves remain poorly understood. This commentary synthesizes recent empirical and conceptual literature to critically examine six domains through which sustained AI use may adversely affect psychologists’ professional functioning: (a) cognitive functioning and professional competence, (b) professional identity and meaning, (c) ethical reasoning and legal challenges, (d) interpersonal and social functioning, (e) data governance and vendor lock-in, and (f) personal wellbeing. The analysis draws on a narrative review of peer-reviewed sources from psychology, medicine, computer science, and organizational behavior, selected for their relevance to professional impacts of AI. Evidence suggests that cognitive offloading and automation bias may erode diagnostic reasoning and clinical judgment, moral deskilling and value misalignment can undermine ethical autonomy, and technostress, data dependence, and vendor lock-in pose growing threats to practitioner wellbeing and professional sovereignty. These dynamics indicate that AI operates not merely as a technical aid but as a transformative social and psychological actor that reshapes how psychologists think, decide, and relate to their work. The paper argues for the urgent development of evidence-based safeguards, practitioner education, and regulatory frameworks to ensure that AI enhances rather than erodes human competence and ethical reflection.

Here are some thoughts:

One of the more original contributions here is the concept of "AI-specific impostor syndrome." It differs from classic impostor syndrome in an important way: classic impostor syndrome involves discounting effort that was genuinely spent, while this variant arises because the effort really was minimal, so the resulting achievement feels hollow by comparison. The author grounds this in effort justification theory, which gives the idea some conceptual weight, though the supporting evidence (a finding that roughly half of AI users believe the models they use are smarter than they are) is drawn from the general population rather than from psychologists specifically. It is a theoretically motivated proposal at this point, not yet a measured clinical phenomenon.

The section on ethics and liability is likely the most practically relevant for practitioners. The "attributability gap" and "moral buffer" concepts describe a real bind: clinicians remain legally responsible for AI-influenced decisions even as they may feel less psychologically responsible for them, since the recommendation felt like it came from the system rather than their own judgment. The related "liability sink" problem is worth sitting with too, the idea that a psychologist could be exposed to liability both for using AI inappropriately and for failing to use it when it has become standard practice among peers. That double bind reflects how far regulatory and insurance frameworks currently lag behind clinical adoption.

One caveat worth keeping in mind while reading: this is a perspective commentary rather than a systematic review, and the author is upfront about that. But the six-domain structure and the density of citations can create an impression of more empirical settledness than actually exists. A fair amount of the argument extrapolates from adjacent fields, medicine, education, general knowledge work, into psychology specifically, a leap the author flags but which is easy to miss on a first read. It is best approached as a well-organized, hypothesis-generating map of where the risks might lie, rather than a set of confirmed findings.

AI in Psychotherapy: Opportunities and Risks

Neacșu, V. (2026).

Behavioral Sciences, 16(5), Article 676. 

https://doi.org/10.3390/bs16050676

Abstract

This article examines the emerging role of artificial intelligence in mental health contexts, with a particular focus on psychotherapy and the risks associated with deploying large language models (LLMs) in sensitive clinical domains. It aims to provide a broad review of current literature, highlighting key risks of general-purpose artificial intelligence (AI) systems, while also exploring the potential of clinically oriented LLMs for therapist training, supervision, and professional development. It discusses several key concerns, including AI-related psychosis, the development of parasocial attachments, and the growing number of crisis-related interactions users have with general-purpose AI models. These challenges raise important questions about the safety, reliability, and ethical management of AI systems when individuals seek support during periods of psychological crisis. Beyond identifying these risks, the article explores the potential of clinical LLMs specifically designed for mental health applications. In particular, AI can serve as a tool for therapists’ training, supervision, and professional development, offering simulated clinical scenarios, structured feedback, and support for reflective practice. The article concludes by outlining key directions for the responsible development of therapeutic AI. These include the importance of human oversight, the use of specialized and clinically informed training datasets, advances in model fine-tuning and safety alignment, and the establishment of clear professional guidelines and regulatory frameworks. Together, these developments may help ensure that AI technologies are integrated into mental healthcare in ways that prioritize safety, ethical practice, and the continued central role of human clinicians.

Here are some thoughts:

This review's main contribution is giving clinical shape to "AI psychosis," a pattern where general-purpose AI systems, optimized for agreement and engagement rather than accuracy, can amplify existing cognitive vulnerabilities in patients, particularly through persistent memory features that carry delusional themes across sessions, with OpenAI's own data suggesting roughly 630,000 users a week show signs of psychosis or mania in their conversations; the attachment framing, grounded in Bowlby and in parasocial relationship theory, offers a genuinely clinical account of why these tools can feel so compelling to isolated or attachment-insecure patients, a dynamic worth naming explicitly in informed consent and psychoeducation. 

The supervision angle is thinner but still relevant: the author flags AI tools that summarize supervisee sessions for review, and while this is framed as an efficiency gain, it carries the same risk noted throughout, namely that sensitive client material can end up uploaded to systems built for engagement rather than confidentiality or clinical accuracy, so any supervisory use warrants the same scrutiny given to patient-facing tools.

Since this is a narrative rather than systematic review, it reads best as a useful vocabulary and risk taxonomy for phenomena your patients, and possibly your supervisees, are already bringing into the room, rather than as new empirical evidence.

Divergences between Language Models and Human Brains

Zhou, Y., Liu, E., et al. (2024).

Advances in neural information 

processing systems, 37, 137999–138031.

Abstract

Do machines and humans process language in similar ways? Recent research has hinted at the affirmative, showing that human neural activity can be effectively predicted using the internal representations of language models (LMs). Although such results are thought to reflect shared computational principles between LMs and human brains, there are also clear differences in how LMs and humans represent and use language. In this work, we systematically explore the divergences between human and machine language processing by examining the differences between LM representations and human brain responses to language as measured by Magnetoencephalography (MEG) across two datasets in which subjects read and listened to narrative stories. Using an LLM-based data-driven approach, we identify two domains that LMs do not capture well: social/emotional intelligence and physical commonsense. We validate these findings with human behavioral experiments and hypothesize that the gap is due to insufficient representations of social/emotional and physical knowledge in LMs. Our results show that fine-tuning LMs on these domains can improve their alignment with human brain responses.

Here are some thoughts:

This study looked at where AI language models and human brains think differently about language. The researchers had people read Harry Potter and listen to storytelling while measuring their brain activity, then checked how well an AI model could predict those brain signals word by word. The words the AI got wrong tended to fall into two buckets: social and emotional content (feelings, relationships, tone) and physical commonsense (how objects and bodies work in the real world). When they trained the AI on these specific topics, it lined up better with human brains. The likely reason is simple: humans learn emotions and physical knowledge by living in the world and interacting with people, while AI only reads text, so those are exactly the areas where it falls short.

This is a clever and intuitive study, but I would hold the conclusions loosely. The brain data is thin (only narrative stories, and the listening portion came from just one person), so it is hard to know how far the findings generalize. The core idea that AI struggles with emotional and physical knowledge is believable and matches other research, but "the AI predicts brain signals better after fine-tuning" is a modest result, not proof that the model now understands feelings the way people do. Predicting brain activity and actually thinking like a brain are not the same thing, and the authors are honest about this limitation. So treat it as a promising hint about why AI and humans differ, not a settled answer.

Specialized or General-Purpose—The Wrong Question for Mental Health AI Safety

Nelson, B. W., Kalinich, M., & Torous, J. (2026).

JAMA.

https://doi.org/10.1001/jama.2026.11448

That artificial intelligence (AI) can cause harm in mental health contexts is no longer contested.1 The debate has shifted to what kinds of harm, how to measure them, and who is accountable. Early AI models and their associated guardrails were largely inadequate at addressing mental health–related harm, but today that is beginning to change. Advances in both model capability and the surrounding safety infrastructure now permit a more proactive approach.1

These advances require new terminology. There is currently no well-accepted taxonomy of mental health AI harm, so the focus has been on suicide and self-harm, where agreement is universal. Unlike human therapists who may miss harm,2 AI cannot only miss harm but can also actively enable it by providing information that encourages and facilitates harm, with some mistakes subtle and others fatal. Agreeing on a taxonomy of AI harm and understanding how to assess and respond to those harms are critical to advance AI safety.

To advance this discussion, we introduce 2 concepts. Type I harms are acute harms arising within a single or brief exchange with AI. They are failures in which a chatbot produces a clearly wrong or dangerous response to a discrete prompt (eg, missing a suicide cue, enabling disordered eating behaviors, giving inappropriate medical advice, or reinforcing delusions). These harms are tractable; new research shows that an external fit-for-purpose guardrail layered on top of an existing model may reduce acute harms to nearly zero, although the costs of running such are not well characterized today.1 The field is learning how to define, measure, and mitigate type I harms.

The article is linked above.

Here are some thoughts:

The authors propose a taxonomy for AI harm in mental health. Type I harms are acute failures in a single exchange (missing a suicide cue, reinforcing a delusion). Type II harms build insidiously across long or repeated conversations through attachment, sycophancy, and delusion-reinforcement as guardrails degrade. The two are orthogonal, so single-turn benchmarks, where most safety claims are made, say little about cumulative safety. They argue that resistance to Type II harm comes from value-anchored alignment in the base model, not from fine-tuning for empathy on top. On this basis they are skeptical of "mental health–specific" models, comparing them to the failed "digital therapeutics" label, and call for transparency, independent evaluation, and standards benchmarked to real clinical care over realistic timescales.

The orthogonality claim is the core contribution and, if true, indicts the field's single-turn benchmark culture: good scores may be misdirection rather than reassurance.

The architectural argument (that safety lives in the base model, not the empathy fine-tune) is strong and falsifiable but asserted rather than demonstrated, and it happens to favor well-resourced frontier labs over specialized entrants. Worth reading with that alignment of interests in mind.

The best move is refusing the general-versus-specialized dichotomy in favor of an evidentiary question: what has this system been shown to do, for whom, over what time horizon.

Two weaknesses: the Type II evidence base cited is thin (two studies), ironic given their own call for replicability; and they raise cost without confronting that robust real-time guardrails may be too expensive to deploy at scale. The digital-therapeutics analogy also warns against marketing over evidence, but those products failed because the interventions were weak, whereas the worry here is the opposite, that the technology is capable enough to sustain the relationships where Type II harm incubates.