OpenAI updates GPT-5 to better handle sensitive mental health chats

OpenAI said on Monday that it has updated the default GPT5 model to change how ChatGPT responds to sensitive mental health conversations, training the system to recognize and intervene when users show signs of severe distress or an unhealthy emotional reliance on the assistant. The company said the revised model, rolled out on October 3, will encourage users to seek contact with real people and professional support rather than serving as a primary source of emotional care.

OpenAI reported that its internal analysis of more than 800 million weekly active users found small but consequential shares of conversations that indicate acute mental health concerns. The company estimated that 0.15 percent of active users in a given week engage in dialogues containing explicit indicators of potential suicidal planning or intent, more than one million people. About 0.07 percent of weekly users showed possible signs of psychosis or mania, and roughly 0.15 percent exhibited signs of heightened emotional reliance on the chatbot. OpenAI also provided message-level figures, reporting that these categories together account for a fraction of total exchanges but translate into millions of messages...