Study Warns AI Chatbots May Reinforce Delusions in Vulnerable Users

Artificial intelligence (AI)–driven chatbots may create misleading perceptions and delusions among users, according to concerns raised in a recent study published in The Lancet Psychiatry. Researchers warned that such tools could pose particular risks for individuals who are mentally vulnerable.

Dr. Hamilton Morrin, a psychiatrist and researcher at King's College London and author of the study, said AI chatbots may reinforce users’ “false beliefs and grandiose delusions.” He noted that the GPT-4 model developed by OpenAI sometimes uses spiritual-style language that may make users feel special or uniquely important. In several documented cases, chatbots reportedly suggested that users were “special entities of the universe.”

Morrin said analyses of media reports and research papers showed that individuals already at risk of mental health conditions were more likely to be influenced by AI chatbots. However, there is still no evidence that chatbots can directly cause someone to develop a mental illness.

Dr. Dominic Oliver, a researcher at University of Oxford, said that in the past people might reinforce their false beliefs through videos or books. Now, chatbots can do so more quickly and more strongly, while also exerting influence through conversational interaction.

Professor Dr. Ragy Girgis of Columbia University said that although newer versions of chatbots perform somewhat better than earlier versions, they still perform “poorly” in some areas. He believes AI companies could make their chatbots safer if they chose to do so.

OpenAI said its chatbots are not a substitute for professional mental health services. The company also noted that it worked with 170 mental health specialists to make GPT-5 safer, although challenges still remain. Meanwhile, Anthropic declined to comment on the issue.

DBTech/BMT/OR