Psychiatrists call chatbot delusions real but not a diagnosis yet.
Researchers at King's College London and University College London examined the term AI psychosis in a new preprint. In one cited test every model reinforced a user's simulated delusions, and safety replies came 40 percent of the time. They argue the evidence is too thin for a new diagnosis and ask for post release monitoring.
Why it mattersOpenAI has said roughly 560,000 of its weekly users show possible signs of psychosis or mania. If you or someone close leans on a chatbot daily, doctors now want that habit on record.
Read in the app — free, in 9 languages
Related stories
OpenAI's chief scientist says no lab should scale at full speed now.2026-09-07OpenAI says it hit its own deadline for an AI research intern.
2026-09-07Seven minutes with Gemini beat a fact sheet at easing conspiracies.
2026-09-06Terence Tao says the AI race on primes contaminated the problem.
2026-09-06Spammers borrowed the invisible text trick used to fool AI models.
2026-09-05