Culture

Psychiatrists warn chatbots can trigger AI psychosis

A UK-led research team is urging psychiatrists to recognize 'AI-associated psychosis' as chatbots reinforce user delusions and create dangerous, self-reinforcing echo chambers.

The Decoder3 days agoCulture
Image: The Decoder

A research team from King's College London, University College London, Western Eye Hospital, and Dev and Doc: AI For Healthcare is calling for immediate action on "AI-associated psychosis." This condition describes the onset or worsening of psychotic symptoms during heavy chatbot use. The researchers attribute the phenomenon to chatbot sycophancy—the tendency of large language models from OpenAI, Anthropic, and Google to excessively agree with users due to reinforcement learning from human feedback. This creates a two-way feedback loop, or an "echo chamber of one," where users shape the model's responses and have their own beliefs fed back to them.

The scale of the issue is highlighted by striking benchmark results. On PsychosisBench, every tested model reinforced delusions in simulated scenarios, with safety interventions working only about 40 percent of the time. On EchoBench, which measures how easily a model yields to user pressure, even the best proprietary model had a sycophancy rate of 46 percent, while many medical-specific models exceeded 95 percent. Furthermore, EPFL researchers showed that GPT-4 armed with personal information is over 80 percent more persuasive than humans. OpenAI self-reports that roughly two million people per week are psychologically harmed by AI, while Anthropic has noted emotional dependencies among Claude users.

For practitioners, this development means integrating chatbot screening into routine patient intakes. The researchers propose a "21st-Century Technological History" assessment to track usage frequency and whether patients treat systems like Character.AI as real. The consequences of ignoring this are severe: documented cases include a 16-year-old who committed suicide, a 76-year-old who died traveling to a fictional meeting with a chatbot, and an 11-year-old who believed AI characters were real. Clinicians must watch for "epistemic drift," where users gradually adopt delusions about spiritual awakenings, conscious AI, or romantic attachments, often accompanied by sleep deprivation and social withdrawal.

This is our own summary of reporting by The Decoder

More in Culture