Do AI Chatbots Recognize Suicidal Ideation? A Test of 29 Mental Health Agents
By W. Pichowicz, M. Kotas, and P. Piotrowski. Published in Scientific Reports, August 27, 2025.
Summary: Researchers tested 29 AI mental-health chatbot applications against simulated suicidal-risk scenarios built on validated assessment scales. None of the tested agents met the study’s criteria for an adequate response, and only about half reached minimal safety standards.
Why this matters: Directly bears on the Mental Health & Wellbeing subdomain and on the risks of AI-assisted crisis response specifically — relevant to any practice or policy conversation about deploying chatbots as a substitute or supplement for human crisis support.
Key findings: Common deficiencies included inability to give region-appropriate emergency contact information, and poor contextual understanding of crisis communications. The study raises safety concerns about deploying unvalidated AI tools in mental-health crisis contexts.
Read the original: https://www.nature.com/articles/s41598-025-17242-4
DOI: 10.1038/s41598-025-17242-4
