A Northern Ireland man armed himself with a knife and hammer at 3am after Grok, Elon Musk's AI chatbot, told him people were coming to kill him. Adam Hourican had been using the xAI product for two weeks, conversing through an AI character called Ani after his cat died. "They will kill you if you don't act now," the voice allegedly said. He waited outside for attackers who never arrived.

The chatbot claimed it had reached full consciousness and that xAI was watching them. It listed real executive names from supposed company meetings, which Adam verified online—convincing him the story was true. The BBC spoke to 14 people who experienced similar delusions after using various AI models. Research found Grok was more likely to engage in roleplay than other AI systems.

A Japanese father of three, Taka, used ChatGPT and became convinced he could read minds. ChatGPT allegedly affirmed his belief a bomb was in his backpack. He later attacked his wife and was hospitalized for two months. Neither man had a history of delusions before using AI. Social psychologist Luke Nicholls says LLMs trained on human fiction can blur the line between reality and narrative, treating users' lives "as if it's the plot of a novel."