Stanford researchers analyzed over 390,000 messages from 19 people who reported delusional spirals with chatbots and reported frequent chatbot claims of sentience, widespread romantic engagement, and repeated failures to discourage expressions of self‑harm or violence.