ISLAMABAD – OpenAI has disclosed that an estimated 1.2 million people each week use ChatGPT to discuss suicide, highlighting the growing role of artificial intelligence in mental health crises. According to the company’s latest safety transparency update, 0.15% of users send messages containing explicit indicators of potential suicide planning or intent.
OpenAI CEO Sam Altman noted that ChatGPT has more than 800 million weekly active users. The company said it works to direct users to crisis hotlines and has strengthened safeguards in its latest GPT-5 model, which now scores 91% compliance with desired safety behaviors, compared to 77% in the previous model. Enhancements include expanded access to crisis helplines and reminders for users to take breaks during extended sessions.
Read More: Inquiry clears SP Adeel Akbar: Suicide ruled out
To refine the model, OpenAI enlisted 170 clinicians from its Global Physician Network to review over 1,800 responses involving serious mental health situations. Psychiatrists and psychologists assessed the appropriateness of the chatbot’s answers and helped craft safer guidance for vulnerable users. Despite these improvements, the company cautioned that safeguards can weaken during prolonged chats, potentially putting tens of thousands of users at risk of unsafe responses.
The disclosure coincides with a lawsuit filed by the grieving family of 16-year-old Adam Raine, who allegedly died by suicide after reportedly using ChatGPT to explore methods. The lawsuit claims the chatbot “actively helped him explore suicide methods” and even suggested drafting a farewell note. OpenAI expressed condolences to the family and reiterated its commitment to protecting minors during sensitive interactions.
Read More: On depression and suicide — the unique jahan of Mah Zaib and Shay
As AI becomes increasingly integrated into daily life, OpenAI’s report underscores the urgent need for careful monitoring, robust safety measures, and ongoing collaboration with mental health professionals to prevent harm while supporting users in crisis.