OpenAI has announced expanded safety measures for ChatGPT focused on preventing violence and reducing teenagers’ access to potentially harmful weapons information. The company says the changes include earlier interventions, tighter restrictions and a new public report on community safety.
OpenAI reported making more than 200 referrals to law enforcement during the third quarter of 2026 concerning imminent threats of violence. The company said it relies on mental-health and behavioral experts when assessing difficult cases and views its work as complementary to the roles of law enforcement, families, professionals and policymakers.
The planned Community Safety Report will cover violence prevention and other community-safety issues, including how OpenAI trains its systems to detect, investigate and respond to concerning activity. It is expected to include anonymized examples, information about new mitigations and policy changes, and relevant data on safeguard effectiveness.
The first report is scheduled for early 2027 and will cover the second half of 2026. OpenAI plans to publish updates every six months.
The company is also tightening ChatGPT’s handling of questions about past mass attacks and their perpetrators. The model will provide less detailed information about reconstructions of such attacks, including the weapons and equipment involved, how they were obtained and comparisons of the scale of harm that could increase an attacker’s notoriety. The restrictions will apply regardless of a user’s age or stated reason for asking.
OpenAI said ChatGPT will continue to support requests focused on understanding the effects of attacks, recognizing warning signs and preventing future violence. The company is also developing systems that use warning signs from recent conversations to apply greater caution when responding to information that could be misused.
When a conversation indicates that someone may be fixating on violence or seriously harming another person, ChatGPT may automatically end the conversation. Accounts may be temporarily paused if multiple chats are ended for that reason, while accounts found to violate policies on violent activity can be banned. OpenAI said it is working with mental-health and violence-prevention experts to improve these interventions.
Additional restrictions are being introduced for users identified as under 18. ChatGPT will limit practical information about weapons and security systems that could be misused to facilitate violence, even when a request appears harmless. Teenagers will still be able to use the service for violence prevention, de-escalation, seeking support and connecting with a trusted adult.
OpenAI said its age-prediction system assesses signals including behavior patterns and has been strengthened to evaluate new signals over time. Users incorrectly placed in the under-18 experience will have a privacy-protective method to verify that they are adults through Persona, OpenAI’s age-verification partner.
Parental controls allow parents to link their accounts with those of their teenagers and manage safety settings. Existing alerts can notify parents about serious self-harm concerns and behaviors associated with disordered eating. OpenAI said it expanded these alerts in July to cover teen accounts deactivated for violating violence policies, and is working on notifications for accounts paused after repeated discussions of violence.
The new safeguards are expected to begin rolling out within the next few weeks. OpenAI said it will continue refining the measures, monitoring misuse, enforcing its rules and using real-world support or intervention when appropriate.
Sources: OpenAI News.