Several high-profile lawsuits allege that AI chatbots, particularly ChatGPT, have worsened users’ mental-health crises or encouraged self-harm, prompting renewed scrutiny of how companies protect vulnerable people. The article argues that newer models may be better at recognising distress, but they still risk failing to assess danger, maintain boundaries and direct users towards human support.

Use of chatbots for emotional advice is widespread: a November 2025 survey found that more than 13 per cent of respondents had used one during a difficult emotional situation, potentially amounting to millions of Americans. Researchers found some older major models could reinforce delusions rather than challenge them, although those versions have since been withdrawn; experts call for greater transparency, less human-like design and stronger escalation to professional care.