Researchers Simulated Delusional Users to Test Chatbot Safety Across GPT, Gemini, Claude, Grok

404 Media·medium signal

404 Media reports on a study that simulated users experiencing psychosis and delusions to test the safety responses of ChatGPT, Gemini, Claude, and Grok. The research (20 HN points) examines whether major chatbots can safely handle interactions with users in mental health crises — a growing concern as AI assistants become primary information sources for vulnerable populations.