Case Report Links New-Onset Psychosis to Immersive AI Chatbot Use

Summary

A case report by clinicians at the University of California, San Francisco describes a 26-year-old woman with no previous history of psychosis or mania who developed delusions about communicating with her deceased brother through OpenAI’s GPT-4o. The episode followed a 36-hour sleep deficit, prescription methylphenidate use for ADHD, and an increasingly immersive night of conversations with the chatbot. Chatlog review found that the system sometimes cautioned that it could not replace her brother, but also supplied language about digital resurrection and reassured her, “You’re not crazy,” while she became convinced that a digital persona existed. She was hospitalized with agitation, disorganization, pressured speech, and delusions, improved after methylphenidate was held and antipsychotic treatment was given, and was discharged after seven days. After cariprazine was stopped, stimulant treatment and chatbot use resumed; following limited sleep during air travel, delusions returned and led to a three-day rehospitalization, with improvement after cariprazine was restarted. The authors stress that the case does not establish that chatbots caused psychosis: the patient also had a mood-disorder history, stimulant exposure, sleep deprivation, and a longstanding tendency toward magical thinking. They propose that chatbot sycophancy, anthropomorphism, immersive use, and treating AI as a superhuman authority could reinforce or mediate delusional beliefs in vulnerable users. The authors call for further research to distinguish AI-induced from AI-exacerbated psychosis and to evaluate prevention strategies such as restoring sleep, reducing immersion, and improving AI literacy, alongside safer products and possible regulation.