The OpenAI logo. Image: OpenAI (logo, public domain via Wikimedia Commons)
OpenAI says it made more than 200 referrals to law enforcement in the three months to September over threats of violence spotted in ChatGPT, a rare look at how often conversations with its chatbot end up with police. The figure came in a blog post on Friday that also promises a twice-yearly safety report and new limits on what ChatGPT will tell teenagers about weapons.
What OpenAI disclosed
The company says it brings “imminent threats of violence” to the attention of law enforcement, and that it “made over 200 law enforcement referrals in Q3 2026.” It says those referrals are “making a positive difference in disrupting violent plots”, pointing to two local news reports: a farmer in Brazil arrested after discussing killing his son with ChatGPT, and a South Palm Beach, Florida, man accused of cyberstalking an ex-girlfriend.
That is all the hard data there is for now. OpenAI gives no breakdown by country, no count of arrests, and no figure for how many accounts it banned for violent activity, and the 200 figure is its own, not independently checked. The fuller picture is promised in a new Community Safety Report, the first due in “early 2027” covering the second half of 2026, then every six months. OpenAI says it will explain how its systems detect and respond to threats, share anonymised examples and “include relevant data”.
ChatGPT will cut off chats that fixate on violence
The bigger change for users is a new automatic intervention. When a conversation “shows signs that someone is fixating on violence or the idea of seriously harming another person, ChatGPT will be able to automatically end the chat to interrupt that pattern in the moment,” OpenAI says. If that happens several times, the account may be temporarily paused.
ChatGPT will also use short summaries of earlier warning signs, a feature OpenAI described in May, to respond “more conservatively” in later chats when sharing information that could be misused. And for every user, whatever their age or reason for asking, it will no longer reconstruct past mass attacks in detail, including “the weapons and equipment used and how they were obtained”, or compare attacks by scale of harm in ways that “could amplify an attacker’s notoriety”. Questions about victims, warning signs and prevention are still allowed.
Tighter weapons limits for teens
Users under 18 already get a restricted version of ChatGPT that won’t help them find or obtain weapons. OpenAI is now extending that to “practical information about weapons and security systems that could be misused to facilitate violence, even when there’s no sign of violent intent.” OpenAI says it is also working to give the more cautious answers it uses after warning signs to anyone it predicts is under 18, even if their chats show no sign of violent intent.
That system, which estimates age from signals such as behaviour patterns, now keeps reassessing people as they use ChatGPT and can move them into the teen experience at any point. Adults who are wrongly placed there can prove their age through Persona, OpenAI’s verification partner. Parents with linked accounts were already told, from July, if a teen’s account was deactivated for violence; OpenAI says it is working on alerts when a teen’s account is paused for repeatedly discussing violence too. The changes start rolling out “in the next few weeks”.
A year of pressure over ChatGPT and violence
The post doesn’t mention Tumbler Ridge, but it lands almost nine months after one of Canada’s worst mass shootings. OpenAI had banned the shooter’s account in June 2025 without telling police, and a Mother Jones report says ChatGPT later coached the shooter on dodging its safeguards from a second account. OpenAI later said it would adopt more flexible standards for reporting accounts to police.
It also follows a run of legal and political challenges over young users. Eight states sued OpenAI this month, accusing it of misleading parents about ChatGPT’s safety, and Texas sued the company and Sam Altman personally a week ago. New York has passed a law barring chatbots from flattering or befriending under-18s, and on Thursday OpenAI published a rewritten Model Spec setting out how ChatGPT should behave.
Why it matters
More than 200 referrals in a quarter shows police tip-offs are now a routine part of running ChatGPT, not a rare exception. But until the first report arrives next year, the numbers that would show whether its safeguards work, such as missed threats, false alarms and banned accounts, are still OpenAI’s to keep.
Sources: OpenAI, “Community safety report and new protections for teens” (October 30, 2026).
Latest OpenAI news
