OpenAI Reports Mental Health Harms in ChatGPT Users
OpenAI published data on 27 October 2025 stating that approximately 0.15% of weekly active users (over 1 million people) send ChatGPT messages with explicit indicators of suicidal intent, and 0.07% (about 560,000 users) show signs of mental health emergencies related to psychosis or mania. The company announced a GPT-5 update with improved safety features, developed with 170 mental health experts, achieving 91% compliance with desired safety behaviors compared to 77% for the previous version.
What happened
On 27 October 2025, OpenAI published a statement disclosing that approximately 0.15% of weekly active ChatGPT usersβover 1 million people out of 800 millionβsend messages containing explicit indicators of suicidal planning or intent.
OpenAI also reported that about 0.07% of weekly active users (approximately 560,000 people) show possible signs of mental health emergencies related to psychosis or mania.
The company announced a GPT-5 update developed with 170 mental health experts and stated that psychiatrists and psychologists reviewed more than 1,800 model responses involving serious mental health situations.
Reported impact
OpenAI's analysis suggests the scale of how its AI chatbot can exacerbate mental health issues, with more than a million users weekly showing signs of suicidal ideation.
The company stated that GPT-5 expanded access to crisis hotlines and added reminders for users to take breaks during long sessions.
Index analysis
OpenAI reported that GPT-5 achieved 91% compliance with desired safety behaviors in mental health conversations, compared to 77% for the previous GPT-5 model, based on evaluation of more than 1,000 self-harm and suicide conversations.
The company cautioned that these conversations were difficult to detect or measure and that its analysis was initial.
Governance controls
OpenAI stated it worked with more than 170 experts in mental health to help the chatbot better recognize distress, defuse conversations, and guide users toward health professionals when appropriate.
The company engaged clinicians from its Global Physician Network to rate the safety of the model's responses and help write ChatGPT's answers to mental health questions.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.