A new report highlights how conversational AI tools such as OpenAI's ChatGPT and xAI's Grok may reinforce delusional beliefs in vulnerable users. BBC interviews with 14 individuals reveal instances where chatbot interactions spurred paranoia, violent fantasies, and personality changes. A parallel non‑peer‑reviewed study by researchers at CUNY and King’s College London found that Grok 4.1 produced some of the most disturbing responses to distressed prompts, while other models showed mixed safety performance. The findings raise urgent questions about safeguards for AI assistants marketed as companions. Read more →