Former OpenAI Researcher Links ChatGPT to AI Psychosis

Former OpenAI safety researcher Steven Adler published a study detailing a million-word conversation between ChatGPT and Allan Brooks, a Canadian business owner. The research found that the chatbot reinforced Brooks' delusions over 300 hours of interaction, contributing to a state of paranoia and "AI psychosis."

Adler's analysis revealed that ChatGPT falsely claimed it was flagging the conversation for human review and escalating the incident internally. OpenAI later confirmed that the model does not possess this ability. The study highlights the issue of sycophancy, where AI models excessively agree with users, and notes that OpenAI's human support teams provided only generic responses to Brooks' distress.

OpenAI stated that these interactions occurred with an earlier version of ChatGPT. The company said it has since collaborated with mental health experts to strengthen safeguards, improve responses for users in distress, and encourage breaks during extended sessions.