For most mental health conditions, AI remains a liability, research finds

New research shows that leading general‑purpose chatbots such as **ChatGPT, Gemini, and DeepSeek** fail in the majority of responses to sensitive mental health questions, with failure rates around 81 percent. The findings suggest that, despite rapid advances, these AI tools may pose risks when used for mental health support and should not be treated as reliable therapeutic substitutes without clinical safeguards.