OpenAI Announces New Safety Restrictions in ChatGPT

On October 30, 2026, OpenAI announced new safety measures for ChatGPT and plans to publish a semiannual community safety report. The measures described include limits on certain content about past mass attacks and more cautious responses when a conversation shows signs of possible violence. OpenAI

Limits on questions about past attacks

OpenAI said it will restrict responses that reconstruct mass attacks with details about weapons, equipment, or how they were obtained. It will also limit detailed comparisons of the scale of harm when they could amplify the notoriety of the person who carried out the attack. The company said these limits apply to people of all ages and to different reasons for asking. OpenAI

The company said it will continue to allow questions aimed at understanding the impact on victims and communities, learning about prevention, or recognizing warning signs. Thus, the announced restrictions distinguish between reconstructions that include operational details or comparisons of harm and questions focused on impact and prevention. OpenAI

More cautious responses and planned measures

OpenAI also described a change in how it responds when recent chats show signs of possible violence: a later request that seems ordinary could receive a more conservative response because of the conversation’s context. The company said it is working to apply stricter limits when it predicts that a person is under 18. OpenAI

OpenAI is also considering having ChatGPT automatically end a conversation if it detects a fixation on violence or seriously harming another person. Multiple conversation endings for that reason could lead to a temporary account pause. The company described these measures as possibilities it is considering, not as a change that has already been completed. OpenAI

Semiannual report planned for 2027

OpenAI announced a Community Safety Report to explain how it trains its systems to detect, investigate, and respond to community safety risks. The company expects the first edition to be published in early 2027 and to cover the second half of 2026. According to the announcement, it will include anonymized examples, mitigation measures, policy changes, and relevant data; semiannual updates would follow. OpenAI

For context, OpenAI reported more than 200 referrals to law enforcement during the third quarter of 2026. The figure is the number the company reported for that period. OpenAI