OpenAI, the maker of ChatGPT, has released new transparency data showing that a small fraction of its users display signs of mental distress, including suicidal ideation or psychosis, during interactions with the AI model.
According to a recent company report, approximately 0.15% of weekly ChatGPT users show indications of suicidal planning or intent, while around *0.07% display language suggesting possible psychosis or manic episodes. The findings were drawn from aggregated, anonymized user data reviewed under the company’s “Sensitive Conversations” monitoring initiative. (OpenAI Blog)
OpenAI stated that the data is being used to improve how ChatGPT identifies and responds to users expressing signs of emotional or psychological crisis. The company emphasized that the goal is to ensure AI systems can provide safe, supportive, and nonjudgmental responses while connecting at-risk users with real-world mental health resources.
“We take these situations very seriously. Our models are designed to recognize language associated with self-harm and other crises and respond with empathetic and evidence-based interventions,” OpenAI said in the report.
The company explained that ChatGPT’s safety mechanisms have been strengthened in collaboration with mental health organizations, including the National Suicide Prevention Lifeline (U.S.) and Samaritans (U.K.), to improve real-time crisis referrals for users in distress.
The report also revealed that while the percentage of affected users remains small, the absolute number is significant given ChatGPT’s global user base, estimated at over 200 million active users.
“Even a small percentage translates into tens of thousands of people each week expressing deep distress,” the company added, calling it a “reminder of the critical importance of responsible AI design.”
Mental health professionals have applauded the move toward transparency but warned that AI platforms must balance user privacy with the ethical need to intervene in emergencies.
Dr. Lena Morales, a digital psychiatry researcher at Stanford University, told The Guardian that such data “offers a vital lens into public mental health trends,” but cautioned that “AI should never replace qualified human support.” (The Guardian)
OpenAI says it will continue refining ChatGPT’s response systems and exploring optional “mental health check-in” features that could allow users to self-identify and access crisis support directly.
“We want ChatGPT to be a safe space — one that encourages people to seek help, not replace it,” the company stated.
The new report comes amid broader industry discussions about the psychological impact of conversational AI and the responsibilities of tech firms to safeguard users during sensitive interactions.
Sources: OpenAI Blog, The Guardian, BBC Tech
