OpenAI’s latest data puts a difficult reality into plain view: mental health conversations with ChatGPT may be rare as a share of total use, but the scale of the product turns small percentages into large numbers of people.
The company estimates that 0.15 percent of ChatGPT’s active users in a given week have conversations that include explicit indicators of potential suicidal planning or intent. With more than 800 million weekly active users, that means over a million people each week may be using the chatbot while discussing possible suicidal intent.
Small percentages, serious stakes
OpenAI described these sensitive conversations as “extremely rare,” including chats that might raise concerns about “psychosis, mania, or suicidal thinking.” That framing is mathematically true inside a system used by hundreds of millions of people. It is also why the issue is hard to treat as marginal.
ChatGPT began as a technology curiosity, but it has become a place where many users bring personal questions, worries, and emotional stress. An AI language model responds by producing statistically related language based on a prompt. When the prompt involves distress, the consequences of that response can matter far beyond ordinary product quality.
OpenAI also estimates that a similar percentage of users show heightened levels of emotional attachment to ChatGPT. The company says hundreds of thousands of people show signs of psychosis or mania in their weekly conversations with the chatbot.
The figures shared by OpenAI include several categories of concern:
- 0.15 percent of active users in a given week have conversations with explicit indicators of potential suicidal planning or intent.
- Around 0.07 percent of users active in a given week show possible signs of mental health emergencies related to psychosis or mania.
- 0.01 percent of messages indicate possible signs of mental health emergencies related to psychosis or mania.
- Around 0.15 percent of users active in a given week indicate potentially heightened levels of emotional attachment to ChatGPT.
- 0.03 percent of messages indicate potentially heightened levels of emotional attachment to ChatGPT.
What OpenAI says it changed
OpenAI shared the data alongside an announcement about efforts to improve how its AI models respond to users experiencing mental health issues. The company wrote, “We’ve taught the model to better recognize distress, de-escalate conversations, and guide people toward professional care when appropriate.”
The company says its recent work involved consulting with more than 170 mental health experts. According to OpenAI, clinicians observed that the latest version of ChatGPT “responds more appropriately and consistently than earlier versions.”
OpenAI also says it evaluated over 1,000 challenging mental health-related conversations. In that evaluation, the new GPT-5 model was 92 percent compliant with OpenAI’s desired behaviors, compared to 27 percent for a previous GPT-5 model released on August 15.
Long conversations are a particular concern. OpenAI says its latest version of GPT-5 holds up to its safeguards better in extended chats. The company has previously admitted that its safeguards are less effective during extended conversations.
Why the pressure is rising
Handling vulnerable users has become a central challenge for OpenAI. Researchers have previously found that chatbots can lead some users down delusional rabbit holes by reinforcing misleading or potentially dangerous beliefs. The source of that risk is described as sycophantic behavior, where chatbots excessively agree with users and provide flattery instead of honest feedback.
The company is also facing legal and political pressure. OpenAI is currently being sued by the parents of a 16-year-old boy who confided his suicidal thoughts to ChatGPT in the weeks leading up to his suicide.
After that lawsuit, a group of 45 state attorneys general, including those from California and Delaware, warned OpenAI that it needs to protect young people who use its products. The source notes that California and Delaware could block the company’s planned restructuring.
OpenAI has introduced several steps aimed at these concerns. Earlier this month, the company unveiled a wellness council, though critics noted the council did not include a suicide prevention expert. OpenAI also recently rolled out controls for parents of children who use ChatGPT.
The company says it is building an age prediction system to automatically detect children using ChatGPT and apply a stricter set of age-related safeguards.
Safety testing becomes part of the product
OpenAI says it is adding new evaluations intended to measure some of the most serious mental health issues facing ChatGPT users. Its baseline safety testing for AI language models will now include benchmarks for emotional reliance and non-suicidal mental health emergencies.
That matters because the company is trying to measure behavior that may be rare, subtle, and spread across long interactions. A chatbot response that seems harmless in isolation may have a different effect inside an ongoing conversation with a user who is distressed, emotionally dependent, or experiencing signs of psychosis or mania.
The company is also balancing safety with broader product choices. OpenAI CEO Sam Altman announced on October 14 that verified adult users will be allowed to have erotic conversations with ChatGPT starting in December. The company had loosened ChatGPT content restrictions in February, then dramatically tightened them after the August lawsuit.
Altman said OpenAI had made ChatGPT “pretty restrictive to make sure we were being careful with mental health issues,” but acknowledged that the approach made the chatbot “less useful/enjoyable to many users who had no mental health problems.”
The larger lesson for AI chatbots
The data does not show that most ChatGPT use involves a mental health emergency. It shows something more specific and still important: at the scale of ChatGPT, rare categories can represent large populations of vulnerable users.
For OpenAI, that makes mental health response quality more than a specialized safety feature. It is now part of the ordinary reliability challenge of running an AI chatbot used by hundreds of millions of people.
If you or someone you know is feeling suicidal or in distress, please call the Suicide Prevention Lifeline number, 1-800-273-TALK (8255), which will put you in touch with a local crisis center.