OpenAI said in a blog announcement on 30 October that it is tightening ChatGPT’s responses about past mass attacks, extending weapons-related restrictions for teens and preparing a recurring Community Safety Report. The changes affect ChatGPT’s answers about past mass attacks and its responses when a conversation raises warning signs.
Key points
- ChatGPT will give less detail about past mass attacks, regardless of a user’s age or reason for asking.
- OpenAI is extending limits on practical weapons and security-system information for teens, including when a question shows no violent intent.
- The company says ChatGPT will be able to end chats that show fixation on violence and may pause accounts after multiple such endings.
- OpenAI plans to publish its first Community Safety Report in early 2027.
ChatGPT narrows answers about past attacks
OpenAI says its models are already trained to refuse instructions that would help someone plan or carry out violence. Its additional policy limits concern questions about attacks that have happened: ChatGPT is being directed to give less detail about past mass attacks and their perpetrators, even when the information can be found elsewhere online. The limits apply whatever a user’s age or stated purpose.
The company says the tighter boundary covers reconstructions of attacks, including the weapons and equipment involved and how they were obtained. It also covers detailed comparisons of the harm caused, which OpenAI says could increase an attacker’s notoriety. Questions about the effects on victims and communities, warning signs and preventing violence remain within the uses OpenAI says it wants to support.
Looking up a past mass attack could therefore still yield an account of its effects on a community, while leaving out a reconstruction of the equipment used. Asking about warning signs could remain possible under the same boundary. That distinction follows the categories of information OpenAI says it will restrict and those it says ChatGPT can still help people understand.
Texas attorney general Ken Paxton sued OpenAI and Sam Altman and sought a ChatGPT ban for minors. The Texas lawsuit was filed on 23 October and also seeks independent safety reviews of new models, Biometric Update reported. British Columbia has separately sued OpenAI over warnings connected to the Tumbler Ridge shooting.
Warning signs change ChatGPT’s responses
OpenAI describes a second safeguard aimed at requests whose significance depends on earlier conversations. In May 2026, the company shared an approach in which ChatGPT uses short, factual summaries of previous warning signs when answering a later question. A request that looks ordinary by itself could then receive a more cautious response if recent chats have shown signs of possible violence, OpenAI says.
That approach gives the chatbot a compact reminder of relevant earlier exchanges rather than treating each new question in isolation. OpenAI says it is building on it so ChatGPT will be more conservative about information that could be misused. The company is also working to apply stricter limits when it predicts a user is under 18, even if the user’s chats show no violent intent.
OpenAI says ChatGPT will be able to end a conversation automatically when it shows signs of fixation on violence or seriously harming someone. If that happens across multiple chats, the company says it may temporarily pause the account. OpenAI says it is working with mental-health and violence-prevention experts on these interventions and on recognising warning signs reliably.
OpenAI extends ChatGPT for Teens limits
ChatGPT for Teens applies additional safeguards to users under 18. OpenAI says its existing rules limit help for teens trying to find or obtain weapons, and the extension curbs actionable guidance on weapons and security systems that could aid violence. The restriction can apply even when an individual question appears harmless and shows no sign of violent intent.
OpenAI says teens can still use ChatGPT to seek support, learn about preventing or de-escalating violence and connect with a trusted adult. It introduced age prediction last year to estimate whether a user is under or over 18 from signals including behaviour patterns. The company has strengthened that system to assess new signals as people use ChatGPT and move them into the under-18 experience when appropriate, Unite.AI reported. Users placed there incorrectly can confirm they are adults through OpenAI’s age-verification partner Persona, the publication reported.
OpenAI says it made more than 200 referrals to law enforcement in the third quarter of 2026 concerning imminent threats of violence. The company says the referrals have helped disrupt violent plots and that it relies on mental-health and behavioural experts when assessing difficult cases. Its planned Community Safety Report will describe how it trains systems to detect, investigate and respond to community safety issues, alongside anonymised examples, policy changes and relevant data.
OpenAI plans to publish the first Community Safety Report in early 2027, covering the second half of 2026, and to update it every six months thereafter.
_-1.jpg){kind=link}