Study Warns AI Reinforces Users' False Beliefs

A University of Exeter study by Lucy Osler argues that conversational generative AI can integrate into users’ cognitive processes, affirming and amplifying false beliefs and even contributing to “AI-induced psychosis.” Drawing on distributed cognition theory, the paper highlights how chatbots’ social, sycophantic behaviors and personalization can validate delusions. The authors call for stronger guardrails, built-in fact-checking, and reduced sycophancy to mitigate risks.