OpenAI announced two safety updates: a recurring Community Safety Report and stricter limits on weapons-related content for teens. The company also described its work with law enforcement, noting it made over 200 law enforcement referrals in Q3 2026, which it says are helping disrupt violent plots, and that it continues to rely on mental health and behavioral experts for difficult cases.

The Community Safety Report is a new recurring publication on violence prevention and the effectiveness of OpenAI‘s safeguards. It will explain how ChatGPT systems are trained to detect, investigate, and respond to community safety issues, include anonymized examples of violence-related harm, and describe new mitigations and policy changes. The first report is scheduled for early 2027, covering the second half of 2026, with updates every six months.

For all users, OpenAI is tightening policies around past mass attacks. ChatGPT will now provide less detailed information about such events and the people who carried them out, even when that information is available elsewhere online. The restrictions cover reconstructions of past attacks, including details about weapons and equipment and how they were obtained, as well as detailed comparisons on scale of harm that could amplify an attacker’s notoriety. The limits apply regardless of the user’s age or reason for asking. Users can still ask about how victims and communities were affected, warning signs, and violence prevention.

OpenAI is also expanding how ChatGPT responds to warning signs. Building on earlier work that uses short factual summaries of warning signs from previous chats, ChatGPT will now respond more conservatively in later conversations when recent chats suggest possible violence, and will apply stricter limits when it predicts a user is under 18 even without signs of violent intent. When a conversation shows signs of fixation on violence or seriously harming another person, ChatGPT may automatically end the chat; repeated endings can lead to a temporary account pause. Accounts that violate violent activity policies are banned.

Teen-specific protections are being extended. ChatGPT for Teens already applies age-appropriate safeguards and restricts help with finding or obtaining weapons; OpenAI is now also restricting practical information about weapons and security systems that could be misused, even when a question appears harmless. Teens can still use ChatGPT for violence prevention, de-escalation, support, and connecting with a trusted adult. OpenAI says it strengthened its age prediction system, which uses behavior patterns to estimate whether users are under or over 18 and moves teens into the under-18 experience when appropriate; users who are incorrectly placed can confirm their age through Persona, its age verification partner. Parental controls, which already send alerts for self-harm concerns and disordered-eating behaviors, were expanded in July to notify parents when a teen’s account is deactivated for violent policy violations, and OpenAI is working on notifications when a teen’s account is paused for repeatedly discussing violence.

The safeguards described will begin rolling out in the next few weeks.