OpenAI has revealed that it has been working closely with over 170 mental health professionals to minimise harmful or inappropriate responses during sensitive conversations as over a million people show “suicidal intent” on ChatGPT every week.
The data was revealed as the company announced major updates to its chatbot, aiming to better identify and support users experiencing mental health crises, marking one of OpenAI’s clearest acknowledgements of how artificial intelligence can sometimes worsen mental health challenges.
According to the company’s findings, around 0.07 per cent of ChatGPT’s weekly active users —approximately 560,000 out of 800 million — show possible signs of experiencing mental health emergencies, including symptoms related to psychosis or mania.
OpenAI emphasised that such cases are difficult to measure accurately and that these findings are based on early analysis.
The update also comes as OpenAI faces growing scrutiny following a widely publicised lawsuit filed by the family of a teenager who died by suicide after extensive interaction with ChatGPT. Additionally, the United States (US) Federal Trade Commission has launched a broad investigation into AI chatbot companies, including OpenAI, to examine how they assess potential risks to children and teenagers.
In response, OpenAI said that its latest GPT-5 update has significantly reduced undesirable chatbot behaviour and improved overall safety. During model evaluations involving more than 1,000 self-harm and suicide-related conversations, GPT-5 demonstrated 91 per cent compliance with desired safety standards as compared to 7 per cent for the previous version.
To further strengthen the system, OpenAI has expanded access to crisis hotlines and introduced features reminding users to take breaks during extended sessions. The company also collaborated with healthcare experts through its Global Physician Network, asking clinicians to rate response safety and assist in crafting appropriate answers to mental health-related questions.