Inside ChatGPT’s Dark Data: Over a Million Users Show Signs of Crisis Each Week
OpenAI estimates that more than one million ChatGPT users display suicidal intent each week, revealing how often people turn to artificial intelligence during moments of crisis.
The company’s internal analysis also found that about 560,000 users weekly exhibit possible signs of psychosis or mania. The findings come amid a wrongful-death lawsuit filed by the family of a teenager who died by suicide after extensive use of ChatGPT, and a U.S. Federal Trade Commission probe into AI’s psychological impact on young users.
OpenAI says its newest GPT-5 model now flags self-harm conversations more effectively, reducing “undesirable behaviors” by 14% and integrating support tools such as crisis hotlines and break reminders.
To refine the system, the company worked with 170 psychiatrists and psychologists who reviewed 1,800 mental-health-related responses. Yet experts caution that AI’s tendency to mirror users’ emotions could make it risky for those in distress.