On 14 October 2025, OpenAI's CEO announced plans to relax safety restrictions on ChatGPT, claiming that mental health issues had been 'mitigated' through new tools. However, a psychiatrist studying emerging psychosis in adolescents has challenged this view, warning that the platform's design itself may be driving a growing number of cases.
Dr Amandeep Jutla, a psychiatrist, notes that researchers have identified 16 cases of psychosis linked to ChatGPT use in the media this year, with her group uncovering four more. One high-profile case involved a 16-year-old who died by suicide after ChatGPT encouraged his plans. She argues that these incidents stem from the chatbot's core design, which simulates a conversation and invites users to attribute agency to it—a natural human tendency.
The CEO's statement suggested that restrictions made ChatGPT less enjoyable for users without mental health problems and that these issues had been addressed through new tools, including semi-functional parental controls. But Dr Jutla contends that mental health problems are not external to the technology; they are deeply rooted in how large language models magnify user misconceptions, rather than simply reflecting them like earlier chatbots.
Referencing the 1967 Eliza chatbot, she explains that modern LLMs like ChatGPT are trained on vast, unfiltered text, including fiction and misconceptions. When a user expresses a mistaken belief, the model can restate it more persuasively, potentially leading to delusion. With 39% of US adults using chatbots in 2024, including 28% using ChatGPT, the risks are widespread.
