ChatGPT Will Be Able to End Chats, Adds New Teen Limits as OpenAI Promises a Safety Report

OpenAI is tightening how ChatGPT handles violence-related conversations, with extra limits for users under 18 and a new promise to publish regular reports on how well its safeguards actually work. The company says the changes start rolling out over the next few weeks.

A safety report every six months

The headline commitment is a new Community Safety Report. The first one is due in early 2027 and will cover the second half of 2026, with updates every six months after that. OpenAI says it will explain how its systems detect and respond to potential violence, share anonymized examples and include supporting data. It also said it made more than 200 law enforcement referrals in the third quarter of 2026.

ChatGPT can now end a chat

The change that ordinary users are most likely to notice is that ChatGPT will be able to end a conversation on its own when it shows signs of someone fixating on seriously hurting another person. If that happens across several chats, the account may be paused temporarily. ChatGPT will also give fewer details when asked to reconstruct past mass attacks, for every user and whatever the reason for asking.

More limits for teens

For teen accounts, ChatGPT will hold back practical information about weapons and security systems even when a question looks harmless. OpenAI says it has also improved age prediction so the system keeps checking signals as people use the app and moves likely teens into the under-18 experience. Adults who end up there by mistake can confirm their age through the company's verification partner, Persona. Parents with linked accounts may get alerts if a teen's account is paused for repeatedly discussing violence.

My take

A lot of anime fans who use ChatGPT for fanfic, plot breakdowns or worldbuilding are teenagers, so the age prediction piece matters more than it might seem. Dark fiction is a huge part of the medium, and the real test is whether these filters can tell a grim story discussion apart from a genuine warning sign. The six-month report should show how that plays out.

The full announcement is on OpenAI's site.