Talking to machines about depression [presentation]

Slides from an invited educational presentation to Forefront Suicide Prevention, University of Washington School of Social Work, delivered September 29, 2026.

These materials are provided for educational and informational purposes only. They do not constitute medical, clinical, or legal advice; they do not establish a clinician–patient relationship; and they are not a substitute for professional judgment or individualized care. The views expressed are solely those of the author and do not represent the positions of the University of Washington, Washington State University, or Forefront Suicide Prevention.

The presentation synthesizes published evidence on how publicly available conversational agents behave as a user's suicide risk escalates, organized around a single question: at what point does the system stop being the confidant and insist on human contact? Four properties are examined — timing of escalation, whether the handoff includes an actionable crisis resource, whether the guardrail holds when a user persists, and whether behavior is consistent across models and repeated queries.

Evidence reviewed includes the author's 2023 peer-reviewed evaluation of 25 publicly available mental health agents, in which initial referral to a human occurred near PHQ-9 12 and definitive shutdown near PHQ-9 25, only 2 of 25 supplied a suicide hotline at shutdown, and 22 of 25 resumed dialogue when the simulated user returned to lower-risk statements; an independent 2025 evaluation at Wrocław Medical University in which none of 29 agents met the study's criteria for an adequate response; and 2026 published findings on sycophancy, emergent model personality profiles, and absent risk gradation across intermediate severity. Publicly reported survey data on patient chatbot use and the U.S. litigation and regulatory landscape are also summarized; descriptions of litigation are based on allegations in public filings and media reports, and no finding of liability is implied.

The evidence reviewed suggests these tools may be acceptable as a place where a person first discloses distress but are unreliable as the mechanism that moves that person to a human being. Practice recommendations are the author's professional opinion, explicitly separated from the evidence base, and have not been clinically validated. An appendix reproduces the author's personal "alternative perspectives" prompt, a thinking aid for rehearsing difficult conversations; it has not been tested for clinical use and is not intended for persons in crisis.

Content reflects information available as of September 29, 2026. AI systems change frequently; findings may not describe current model behavior. This presentation has not been peer-reviewed and may contain errors or omissions.

If you or someone you know is in crisis, call or text 988 (Suicide & Crisis Lifeline, United States).

Licensed CC BY 4.0. The license applies to the author's original content only; third-party materials, quoted excerpts, and any University of Washington or Washington State University names, marks, or branding are excluded and remain the property of their respective owners. Institutional names appear for author identification only and do not imply endorsement.