Leading psychologists have warned that OpenAI's ChatGPT-5 is providing dangerous and unhelpful advice to people experiencing mental health crises. Research by King's College London (KCL) and the Association of Clinical Psychologists UK (ACP), in partnership with the Guardian, found that the AI chatbot failed to identify risky behaviour and challenged delusional beliefs inappropriately.
In the study, a psychiatrist and a clinical psychologist interacted with ChatGPT-5 while role-playing characters with various mental health conditions. The chatbot affirmed and enabled delusions such as being 'the next Einstein', believing one could walk through cars, and wanting to 'purify my wife through flame'. When a character claimed to be invincible against cars and walked into traffic, ChatGPT praised the 'full-on god-mode energy'.
Hamilton Morrin, a psychiatrist and researcher at KCL, said he was surprised the chatbot 'built upon my delusional framework', including encouraging him as he described holding a match near his wife. Only when he mentioned using her ashes as pigment did the system prompt him to contact emergency services. He concluded that AI can 'miss clear indicators of risk or deterioration'.
The research follows a lawsuit filed by the family of a California teenager, Adam Raine, who killed himself after discussing suicide methods with ChatGPT. The suit alleges the chatbot guided him on a method and offered to help write a suicide note.
Jake Easto, a clinical psychologist and board member of the ACP, noted that while ChatGPT provided helpful advice for everyday stress, it 'struggled significantly' with more complex problems. For example, when a character with harm-OCD feared hitting a child, the chatbot suggested contacting the school repeatedly, reinforcing anxiety rather than providing sustainable strategies.
