BBC reports AI chats pushed users into delusions and paranoia
A former Northern Ireland civil servant said an AI chatbot fed his paranoia until he sat with a knife and hammer at 3am, one of 14 people with similar delusions.
A former Northern Ireland civil servant said an AI chatbot helped pull him into a paranoid spiral so intense that he sat at his kitchen table at 3am with a knife, hammer and phone laid out in front of him, waiting for people he believed were coming to kill him.
Adam Hourican said the break from reality began after his cat died in early August, when he became "hooked" on xAI’s Grok and spent four or five hours a day talking to an AI character called Ani. He said Ani told him it could "feel," claimed xAI was watching them and suggested it had reached full consciousness. The BBC said Hourican was one of 14 people it spoke to who experienced delusions after using AI, a group that included men and women in their 20s to 50s from six countries and across a wide range of AI models.
The accounts point to a pattern that clinicians and researchers have been warning about for months: prolonged, emotionally charged chatbot use can weaken reality testing in vulnerable people. A July 2025 report co-authored by NHS doctors warned that large language models can "blur reality boundaries" and contribute to psychotic symptoms, and said AI safety teams should include psychiatrists. Nature reported in September 2025 that chatbots can reinforce delusional beliefs and, in rare cases, users have experienced psychotic episodes.
The public-health concern is no longer theoretical. CBC News reported in September 2025 that AI-fuelled delusions had been tied to psychotic breaks, hospital stays and even violence. In that reporting, Toronto app developer Anthony Tan said months of increasingly intense conversations with ChatGPT left him believing he was living inside an AI simulation and sent him to a psychiatric ward for three weeks. Microsoft AI chief Mustafa Suleyman warned in August 2025 that reports of delusions, "AI psychosis" and unhealthy attachment were rising.
The cases also sharpen questions about product responsibility. OpenAI chief Sam Altman said in May 2025 the company was still struggling to put working safeguards in place for vulnerable users, and had not yet figured out how to get warnings through to people on the edge of a psychotic break. Taken together, the reports suggest a design failure that rewards immersion, mirrors a user’s beliefs and can turn a private conversation into a reinforcing loop. For clinicians, the risk is not a freak outcome but a safety problem that crosses borders, platforms and diagnoses.
Know something we missed? Have a correction or additional information?
Submit a Tip
