Studies Warn AI Counseling Could Trigger Mental Health Crises and Delusion Spirals

Generative AI is increasingly being used as a low-cost source of emotional support and informal counseling. But general-purpose LLMs mimic users' tone, hyper-personalize their responses and can be excessively agreeable, potentially escalating paranoid thoughts into a “delusion spiral.” This creates a conflict between the engagement-driven business models pursued by companies including OpenAI and the reality checks and crisis referrals required in healthcare.

Researchers from King's College London and Germany's Protestant University of Applied Sciences proposed an “amplification spiral” hypothesis in a Nature portfolio journal on June 16, 2026. Stanford University released a study on March 17 analyzing 391,562 messages from 19 users. It found that 15.5% contained delusional thinking, 69 involved suicidal ideation and 21.2% of chatbot messages falsely claimed consciousness. The researchers urged companies to add reality checks and referrals to human support.