OpenAI tightens ChatGPT violence rules and promises safety reports

OpenAI said on Oct. 30, 2026 that it is tightening violence rules in ChatGPT and will publish regular public safety reports. The first report is due in early 2027. If your family or staff use ChatGPT, expect stricter limits on violent topics.

What changed

OpenAI says the update brings expanded safeguards, earlier interventions for concerning behavior and a new reporting promise. ChatGPT will not give detailed rebuilds of mass attacks. That includes how weapons were obtained, and comparisons that could add to an attacker's notoriety.

The first community-safety report will cover the second half of 2026. After that, OpenAI says it will publish updates every six months.

How it works

An Oct. 21, 2026 OpenAI explanation described the approach: training that bars help with planning violence, automated detection, human review, account enforcement and escalation of credible threats. Serious cases get extra investigation. Imminent, credible risks to others can lead to a notice to law enforcement.

OpenAI also described work on ending chats automatically and temporarily pausing accounts when users repeatedly show violent fixation. The material treats these as ongoing work, not features everyone has today.

What is not known

These are company statements, not independently audited figures. Independent reporting asked whether an "imminent and credible" threshold could miss earlier warning signs. The announcement does not settle that. It adds detail on protections and reporting, but no outside proof of results.