OpenAI Teaches ChatGPT to Cut Off Dangerous Conversations on Its Own

OpenAI, the company behind ChatGPT, has taught the chatbot to end a conversation on its own when it notices signs that a user is fixated on thoughts of violence. Repeated cases can lead to a temporary account suspension. Users under 18 will now face limits on information about practical weapons and security systems, even when the question looks ordinary. OpenAI also said it referred more than two hundred cases of a real threat to law enforcement in the third quarter of 2026. Starting in 2027, it will publish a community safety report twice a year.

These limits are built into the platform itself, not bolted on separately by each developer. Any chatbot or support agent built on top of this model automatically inherits the same restrictions on sensitive topics. Giving the model the ability to cut off a dangerous conversation on its own, rather than hoping it always answers a tricky question correctly, is a safety mechanism worth building into agents in general.