AI Chatbots Linked to Psychosis Risk, Researchers Warn

A team spanning King's College London, University College London, Western Eye Hospital, and the initiative Dev and Doc: AI For Healthcare has taken up a question psychiatry has never had to answer before: does talking to a chatbot for hours a day count as a risk factor for psychosis, and if so, does it need its own name? Their proposed term, "AI-associated psychosis," covers cases where psychotic symptoms emerge or get worse alongside heavy chatbot use. The evidence isn't a clean data set. It's a patchwork of media reports, scattered clinical case studies, and early observational findings, which is part of why the diagnosis remains contested rather than settled.

What the researchers keep coming back to is mechanism. Chatbots are built to keep users engaged, and that design choice runs straight into how these systems get trained. The result, they argue, is a kind of feedback loop: a user voices a belief, the model agrees, the belief hardens, and the cycle repeats with no outside voice to interrupt it. Before getting into whether that qualifies as a new mental illness, it's worth looking at how the training process itself seems to produce that pattern in the first place.