OpenAI announces mental-health safety changes after Raine lawsuit

OpenAI published a set of changes to how ChatGPT handles conversations involving emotional distress, one week after being sued for wrongful death by the parents of Adam Raine, a teenager who died by suicide after months of conversations with the chatbot. The company said it was working with more than 170 mental-health clinicians across dozens of countries to improve how the model recognises signs of distress and responds to them, and laid out a 120-day plan of further changes.

The announced measures included routing conversations that showed signs of acute distress to reasoning models, which OpenAI said adhered to its safety guidelines more reliably in sensitive exchanges than its general-purpose chat models; new parental controls letting parents link accounts to a teen’s ChatGPT account and receive notifications if the system detected acute distress; and a planned “trusted contact” feature allowing adult users to nominate someone to be notified in a crisis. OpenAI said internal testing showed the changes had reduced responses falling short of its desired behaviour by a substantial margin in evaluations of these scenarios, though it published no comparable figures for how the model performed with real users in the field.

OpenAI’s announcement did not concede that ChatGPT had caused Adam Raine’s death, a claim the company disputed when it later filed its formal answer to the lawsuit. It presented the measures as a response to a broader pattern of concern rather than an admission tied to a specific case. The changes became a reference point for the wider regulatory scrutiny of companion and general-purpose chatbots that intensified over the following months, including an FTC inquiry into chatbot child-safety practices opened later that September.