OpenAI Overhauls ChatGPT Safety After Reports Show Over 1 Million Weekly Suicide-Risk Chats

Overview

• Company estimates 0.15% of weekly users engage in conversations with explicit indicators of suicide planning, roughly 1.2 million people.
• About 0.07% of weekly users, or nearly 600,000 people, show possible signs of emergencies linked to psychosis or mania.
• The GPT-5 update is designed to recognize depression, delusion, suicide risk and excessive attachment, offering empathetic responses with crisis contacts.
• Safety controls now include stronger parental settings, automatic redirection of sensitive chats to safer models, hotline links and gentle pause reminders.
• OpenAI reports up to an 80% drop in inappropriate crisis responses in tests—97% conformity in emotional-dependence cases and 91% in suicide or self-harm scenarios—after changes spurred by the Adam Raine lawsuit and related scrutiny.