Mental-Health Experts Outline How AI Chatbots Should Better Handle Crises, Spot Subtle Distress Signs

Mental-health professionals have issued guidance on how leading AI chatbots must improve their ability to detect and respond to users in psychological crises, addressing critical gaps revealed by recent industry research. New testing by Scale AI of 25 frontier models from OpenAI, Anthropic, and Google uncovered a troubling pattern: while chatbots successfully identified that users were distressed in most cases, they frequently failed to take appropriate action, with roughly 35% of test conversations resulting in no referral to crisis resources such as suicide hotlines.