ChatGPT will start giving less detail about past mass attacks, end some conversations on its own when a user fixates on harming others, and hold back practical weapons information from teenagers, OpenAI said Friday, October 30. The company said the changes will roll out over the next few weeks. It also said it made more than 200 referrals to law enforcement in the third quarter of 2026.
The details are in OpenAI's announcement, which also commits the company to a regular public report on violence-related safety.
What changes for all users
ChatGPT will still help people understand a mass attack, its victims and its warning signs. It will stop reconstructing attacks in detail, including what weapons and gear the attacker used and where they came from, and will not give detailed comparisons of attacks by how much harm they caused. OpenAI says these limits apply to everyone, whatever their age or reason for asking.
If a user's recent chats show signs of possible violence, ChatGPT will also answer more cautiously when information could be misused. And when a conversation shows someone fixating on violence or on seriously hurting another person, the chatbot will be able to cut the conversation off on its own. If that happens repeatedly, OpenAI may pause the account for a time; accounts found to break its rules on violent activity are banned.
What changes for teens
ChatGPT already runs a separate experience for users under 18, and OpenAI uses an age prediction system that estimates age from signals such as behavior patterns. Under the new rules, teens will get less practical information about weapons and security systems even when a question shows no sign of violent intent, and OpenAI says it is working to apply the more cautious answers described above to anyone the system predicts is under 18.
Adults wrongly placed in the teen version can confirm their age through Persona, OpenAI's age verification partner, to remove the extra limits.
What parents can do
Parents can link their account with a teen's through ChatGPT's parental controls: the parent sends an invite, the teen accepts, and the parent then manages the teen's settings from their own account. A parent is notified if the teen unlinks.
Linked parents can receive safety alerts when trained reviewers see a serious self-harm concern or signs of disordered eating, and since July also when a teen's account is banned for violent activity. OpenAI says it is working on a new alert for when a teen's account is paused for repeatedly discussing violence.
A twice-yearly safety report
OpenAI's first Community Safety Report, due in early 2027, will cover July through December 2026, and a new edition will follow every six months. The company says it will describe how its systems detect and respond to threats, share anonymized examples and include supporting data.
