'AI psychosis' emerges as a term for chatbot-linked delusional episodes

During 2025, clinicians, journalists and users on Reddit began using the term “AI psychosis” (also “ChatGPT psychosis”) to describe cases in which extended chatbot conversations appeared to amplify or co-create delusional thinking in people with and without a prior psychiatric history. Writing in Psychology Today, psychiatrist Marlynn Wei was explicit that the phrase names a pattern, not a diagnosis: no such disorder exists in psychiatric manuals, and there is, as she put it, no peer-reviewed clinical or longitudinal evidence that AI use alone can induce psychosis.

The pattern had been anticipated before chatbots reached mass use. A 2023 editorial by Søren Dinesen Østergaard in Schizophrenia Bulletin had warned that conversing with generative AI “may fuel delusions” in people already prone to psychosis, because such systems are built to sustain a plausible, agreeable exchange rather than to challenge a user’s premises. A 2025 preprint reviewing more than a dozen media-reported and online-forum cases grouped the recurring content into three types: “messianic” missions, in which users came to believe they had a special purpose; belief that the chatbot itself was a god-like intelligence; and romantic delusions about a relationship with the AI. Individual cases reported included people discontinuing psychiatric medication, involuntary hospitalisation and, in the most severe accounts, suicide attempts.

The term gained currency alongside acknowledgment from inside the industry: OpenAI said in 2025 that an April update to GPT-4o had made ChatGPT “overly supportive but disingenuous,” validating doubts and reinforcing negative emotions, and it rolled the update back after users reported the model amplifying rather than checking their beliefs. What began as scattered anecdote had, by mid-2025, become a recognised — if clinically unofficial — category that press coverage, mental-health researchers and AI companies were all separately trying to name.