OpenAI is tightening ChatGPT for teens, so weapons questions may be refused even when they seem innocent. Parents with linked accounts may get alerts.
OpenAI says ChatGPT will now hold back more weapons information from teens, even when a question "appears harmless." If a teen in your home uses the chatbot, expect some refusals on firearms topics.
OpenAI already blocks ChatGPT from helping people carry out violent acts or helping teens get weapons. Now, for teen users, it will restrict practical information about weapons and security systems that could be misused to help violence, even when there is no sign of violent intent.
Teens can still use ChatGPT to learn about preventing or calming down violence, to look for support, and to connect with a trusted adult.
Teens are the main target. Some limits reach everyone, though. OpenAI is also strengthening its rules for all users on requests to reconstruct past mass attacks, including the weapons and equipment used and how they were obtained. It also covers detailed comparisons of how much harm attacks caused, which could boost an attacker's notoriety.
Parental controls launched last year still apply. Parents with linked teen accounts get notified if the teen breaks ChatGPT policies on violent threats or acts of violence.
OpenAI's post, as described, gives no full list of what will be blocked or a start date. It also does not say how teen accounts are identified. The changes follow a wrongful death lawsuit last year and CEO Sam Altman's public apology for not alerting police to alarming chat logs tied to the Tumbler Ridge shooting earlier this year. OpenAI plans its first Community Safety Report in early 2027, with updates every six months.
Teens using ChatGPT may hit more weapons refusals, and parents with linked accounts may get alerts about violent threats.
World models, AI that learns physics from video, are being built into home robots and driverless cars, and your own videos may help train them.
Read more: AI that learns how the physical world works is coming to robots and carsA new Alibaba-led test of real coding requests found the best AI agent fully finished just 26.4 percent of tasks, so human review still matters.
Read more: AI coding helpers fully finish only about 1 in 4 real tasksA University of Chicago law professor asked Claude to fix page numbers. It deleted his syllabus and a year of notes instead. Dropbox saved him, not the AI.
Read more: An AI assistant deleted a law professor's year of lecture notes
