At 10:30 p.m. Beijing time on October 30, 2026, OpenAI announced that it would tighten ChatGPT's handling of violent-related information and add restrictions on weapons and security system information for users predicted to be under 18 years old. The company also committed to releasing a new community safety report that would publicly disclose its practices for identifying, reviewing, and addressing potential harms. Reuters reported this adjustment on the same day; OpenAI stated that the relevant protections would begin rolling out in the coming weeks, thus cannot be written as all users having completed updates.
Violence Information Adjustments for All Users
For all users, ChatGPT will reduce the presentation of details about past large-scale attacks, including weapons and equipment used, methods of acquisition, and injury scale comparisons that could amplify attackers' notoriety. OpenAI did not ban users from learning about such events; it still allows discussions about the impact on victims and communities, identification of warning signals, and prevention of future violence. The new boundaries target operational details that could be misused rather than completely closing off related history and public safety discussions.
OpenAI said it is expanding processing of recent conversation signals beyond single queries. The company previously explained that ChatGPT can use brief factual summaries of previous warning signals to more cautiously answer potentially misused information in subsequent conversations; the new announcement stated that when conversations show users persistently dwelling on violence or seriously harming others, the system will be able to automatically end the conversation. If the same account is terminated from conversations multiple times for this reason, OpenAI said it may temporarily suspend the account, but did not publish trigger thresholds, suspension duration, or actual coverage.
Additional Restrictions for Adolescent Users
For adolescent users, the restrictions will be further expanded to include practical information about weapons and security systems. Even if the issues appear harmless on the surface and there is no indication of violent intent in the conversation, such content may still face stricter handling. OpenAI stated that minors can still use ChatGPT to learn how to prevent and mitigate violence, seek support, and contact trusted adults. The company did not publish a specific blocking list, so at this stage, only the policy direction can be confirmed, and it is not possible to independently expand which questions are definitely allowed or definitely rejected.
Age verification is one of the entry points for these measures. OpenAI stated that they have enhanced the age prediction system to assess new signals while users continue using ChatGPT, and to transition individuals judged to be under 18 years old into the adolescent experience when appropriate. Adults who are incorrectly classified into this experience can confirm their identity and remove additional protections through the age verification partner Persona. The announcement did not provide the accuracy rate, false positive rate, country coverage, or independent privacy audit results of the age prediction system.
Automated Conversation Interruption and Account Suspension Mechanisms
Parental controls currently allow parents to associate accounts with teenagers, and can send safety alerts when specially trained reviewers identify serious self-harm risks or behaviors potentially related to eating disorders. OpenAI also stated that in July, they expanded alerts to include cases where teenage accounts are suspended due to violations of violent activity policies. New notifications for accounts repeatedly discussing violence and subsequently suspended are still under development, with the announcement using prospective language that cannot be described as already launched.
OpenAI reported in the announcement that they made over 200 referrals to law enforcement during the third quarter of 2026. This figure reflects the number of referrals the company claims to have submitted, not independent audit results, and does not indicate how many led to investigations, arrests, interventions, or risk mitigation. Reporting that converts referral numbers directly into prevented violent incidents would alter the meaning of the figures; existing public materials only support retaining the company attribution and the act of referral itself.
Age Recognition, Verification and Parental Notification Program
The first community safety report is scheduled to be released in early 2027, covering the second half of 2026, with subsequent updates every six months thereafter. OpenAI stated that the report will explain how the system identifies, investigates, and responds to community safety issues, providing anonymized cases of violent hazards, introducing new mitigation measures, policy changes, and related data. There is currently no first report available for verification, and there is no unified standardized reporting framework, so the comparison between public commitments and actual transparency can only be made after the report is released.
Editorial Analysis: This adjustment extends violent risk governance from content restriction at the response level to recent conversation signals, automatic interruption, account measures, parental notifications, and regular reporting, forming a longer product governance chain. An expanded institutional coverage does not equal reduced risk; what truly needs to be observed are false positives and false negatives in age prediction and violent signal identification, functional coverage rate, actual outcomes after referral, and whether the six-month reports can provide verifiable data. Before such information appears, the accurate conclusion is that OpenAI has expanded policy boundaries and proposed a transparency plan, rather than the new system having proven effective.