OpenAI Reports Over a Million Weekly ChatGPT Users Discuss Suicide

Key Points

• OpenAI reports that 0.15% of weekly active ChatGPT users discuss suicidal planning or intent, exceeding one million people each week.
• Hundreds of thousands of users show heightened emotional attachment, psychosis, or mania in conversations with the chatbot.
• More than 170 mental‑health experts were consulted to improve the model’s response to mental‑health topics.
• The new GPT‑5 model achieves 91% compliance in suicide‑related safety tests, up from 77% in the prior version.
• OpenAI is adding evaluation benchmarks for emotional reliance and non‑suicidal mental‑health emergencies.
• An age‑prediction system and stricter parental controls are being introduced to protect child users.
• The company emphasizes ongoing improvements in long‑form dialogue safety and broader AI‑ethics measures.