The Pulse
OpenAI Reports More Than 200 Violence Referrals and Tightens Teen Safeguards
OpenAI says it made more than 200 law-enforcement referrals about imminent threats of violence in the third quarter of 2026. The company also announced tighter limits on weapons information for teens and new ChatGPT interventions for conver
AI.info Team · Written with AI · fact-checked ·
OpenAI says it made more than 200 referrals to law enforcement over imminent threats of violence in the third quarter of 2026, a figure it disclosed on October 30 alongside new restrictions for teen ChatGPT users and changes to how the service handles conversations about violence. The company says the referrals helped disrupt violent plots, but its announcement does not give a breakdown of cases, referral outcomes or the number of conversations reviewed.
“We made over 200 law enforcement referrals in Q3 2026.”
OpenAI
OpenAI plans its first safety report for early 2027
The company says its first Community Safety Report will cover the second half of 2026 and appear in early 2027, with updates every six months. OpenAI says the report will describe how its systems detect, investigate and respond to community-safety concerns, and include anonymized examples, policy changes and data about the safeguards’ effectiveness.
The planned report would add detail to a referral count that currently offers little visibility into how OpenAI identifies a credible threat or what happens after it contacts authorities. The announcement does not define what qualifies for referral, say how many cases led to an intervention, or provide a comparison with earlier quarters. OpenAI says it works with mental-health and behavioral experts to assess difficult cases.
ChatGPT will limit details about past attacks
OpenAI says it is tightening responses to questions about past mass attacks, including requests for details about weapons and equipment, how attackers obtained them, and comparisons of the scale of harm. The new limits apply to users of any age, even when a question does not appear to express violent intent.
The company says ChatGPT can still help users understand the effects of attacks on victims and communities, recognize warning signs and learn about prevention. The distinction OpenAI describes is between discussing an event and providing a detailed reconstruction that could help someone reproduce it or amplify an attacker’s notoriety.
Teen accounts face broader weapons restrictions
For users under 18, OpenAI plans to restrict practical information about weapons and security systems that could be misused to facilitate violence, including when a prompt appears harmless. The company says ChatGPT for Teens will still support requests to prevent or de-escalate violence, seek help and connect with a trusted adult.
OpenAI also says its age-prediction system will keep assessing new signals as people use ChatGPT and move users it estimates are under 18 into the teen experience. Adults placed there by mistake can confirm their age through Persona, OpenAI’s age-verification partner, and have the additional teen protections removed.
Repeated violent fixation could end a chat
OpenAI says ChatGPT will be able to end a conversation when it detects signs that someone is fixating on violence or seriously harming another person. If the service ends multiple chats for that reason, the company may temporarily pause the account. These interventions are described as upcoming capabilities, not as measures already operating across all accounts.
Parents with linked accounts already receive some safety alerts, and OpenAI says it expanded those alerts in July to cover teen accounts deactivated for violating its violence policies. The company is working on a separate notification for parents if it pauses a teen’s account after repeated violence-related discussions.
OpenAI says the new safeguards will begin rolling out in the next few weeks. The planned report may give the public a clearer way to assess the company’s threat referrals and interventions; for now, the disclosed figure is a count of referrals, not a measure of how often ChatGPT encounters violent threats or how many cases authorities resolve.