ChatGPT limits weapons information for teens

Teens get no practical weapons details, even without violent intent

The change was announced on October 30, 2026, starts rolling out over the next few weeks, and covers every user whom age prediction places in the under-18 experience, even when a prompt looks harmless. OpenAI paired the new limits with a commitment to publish a Community Safety Report.

Reuters framed the announcement as part of a wider debate over how chatbots can harm young people and what AI companies should do to catch threats. Reuters also pointed to OpenAI's first report on teen use, released earlier in October, which put average teen time on ChatGPT at under 15 minutes a day.

Weapons and security systems are both covered

OpenAI already barred help with planning violence and limited help for teens trying to find or obtain weapons, and it now also restricts practical details that could be misused for violence. The rule covers two kinds of information: weapons and security systems. OpenAI describes the change as an extension of protections it already had, not a new policy from scratch.

Teens can still get help preventing violence

Teens can keep using ChatGPT to help prevent and de-escalate violence, seek support and reach a trusted adult. Only information that could be misused to carry out violence is limited. Learning to recognize warning signs and to prevent future violence stays open to users of every age.

Age prediction keeps checking while people use ChatGPT

Because the check repeats during use instead of happening once at sign-up, an account can move into the under-18 experience mid-use, and adults placed there by mistake can confirm their age through Persona. Age prediction estimates whether a user is under or over 18 from signals such as behavior patterns. OpenAI launched it in 2025 and has now strengthened it to keep weighing new signals as people use ChatGPT.

The new weapons limits apply to the users it places in the under-18 experience (related: ChatGPT for Teens: Age Prediction and Study Mode). OpenAI says the Persona check is fast, protects privacy and is available at any time.

Chats fixated on violence will end automatically for everyone

When a conversation shows someone fixating on violence or on seriously harming another person, ChatGPT will end it regardless of age, and warning signs in recent chats will now make its answers more cautious. OpenAI is building these automated interventions with mental health and violence prevention experts. Ending the chat is meant to break the pattern in the moment.

A request that looks ordinary can mean something else after an earlier conversation. In May, OpenAI explained that ChatGPT keeps short, factual summaries of earlier warning signs so it can answer more carefully in a later chat. Now, when recent chats show warning signs of possible violence, it is also more conservative with information that could be misused.

Repeated endings can pause an account

An account whose chats ChatGPT ends several times over a fixation on violence may be paused, and accounts found to break the rules on violent activity are banned. One ended chat does not trigger a pause by itself. Bans are not new either: OpenAI already bans accounts it finds have broken its rules on violent activity.

Parents may soon hear about paused teen accounts

OpenAI is working on alerts for parents with linked accounts when a teen account is paused for repeatedly discussing violence. Parental controls let a parent link their ChatGPT account to a teen's and manage safety settings together.

Linked parents can already get safety alerts when trained reviewers spot a serious self-harm concern or behaviors tied to disordered eating. Since July, they are also told when a teen account is deactivated for breaking the usage policies on violence.

Details on past mass attacks are now limited for adults too

ChatGPT will hold back the weapons, equipment, sourcing and scale-of-harm details of past mass attacks even when they are available elsewhere online, whatever the user's age or reason for asking. Comparisons of the scale of harm are on the list because they could add to an attacker's notoriety.

Asking what happened, or how victims and communities were affected, still works. So does learning to recognize warning signs and prevent future violence.

OpenAI made over 200 law enforcement referrals in Q3 2026

The referrals concerned imminent threats of violence, difficult cases were assessed with mental health and behavioral experts, and OpenAI says the referrals helped disrupt violent plots. OpenAI says AI companies have a distinct role alongside police, families, mental health professionals and policymakers. It also commits to working directly with the people and agencies best placed to act in the real world.

A Community Safety Report arrives in early 2027

The report will explain how OpenAI trains its systems to detect and respond to safety issues, with anonymized examples, new mitigations and supporting data, starting with the second half of 2026. The first edition comes out in early 2027, and new editions follow every six months. It will also describe policy changes and how well the safeguards are working.

Only linked parents will hear when a teen account is paused

Teen limits need no setup from parents because age prediction applies them, but alerts reach only parents who linked accounts through parental controls, both the deactivation alerts in place since July and the pause alerts still in the works. The weapons limits apply without anyone turning on a setting.

Alerts work the other way around. Without a linked account, a parent will not hear that a teen account was deactivated, and the planned alerts for paused accounts are set to reach linked parents only.