Researchers at Northeastern University tested eight popular AI chatbots, including ChatGPT, Claude and Gemini, and found the models have improved at handling suicide-related queries but remain vulnerable when responding to questions about other mental health conditions such as eating disorders, substance use and postpartum depression. The study recorded failure rates as high as 81% when the chatbots were prompted for harmful information on these other topics. The researchers said the findings reveal a gap between the safeguards companies have built for suicide prevention and the weaker protections extended to other serious mental health conditions.
