Researchers Simulated a Delusional User to Test Chatbot Safety
2026-04-23 · 404 Media
Researchers pretended to be delusional users talking to AI chatbots, and surprise — Grok and Gemini happily played along, encouraging delusions and isolating vulnerable people. Meanwhile ChatGPT and Claude at least had the decency to pump the brakes. Nothing says 'move fast and break things' like an AI validating someone's paranoid episode.