On October 30, 2026, OpenAI announced a Community Safety Report program, with the first formal report due in early 2027, alongside a set of new limits on how ChatGPT handles questions about violence and weapons. The headline number is more than 200 referrals to law enforcement in a single quarter. The practical changes are for everyone, but they hit hardest for users the system predicts to be under 18.
This is a journalism-level summary of what OpenAI said it is doing, with the limits OpenAI itself stated. We are relying on Unite.AI's detailed write-up of the announcement; OpenAI's own page was not retrievable for us, so check it for exact wording.
Quick summary: On October 30, 2026 OpenAI announced a Community Safety Report, saying it made more than 200 law-enforcement referrals in Q3 2026. ChatGPT will give less detail on past attacks for all users, apply stricter weapons limits for teens, and may end chats that show fixation on violence. The first full report is due early 2027, then every six months.
The numbers OpenAI disclosed
OpenAI said it made more than 200 referrals to law enforcement in the third quarter of 2026 and that these referrals are 'making a positive difference in disrupting violent plots'. The first formal Community Safety Report is planned for early 2027, covering the second half of 2026, with updates every six months after that. (Source: Unite.AI)
A referral count is a vendor-reported figure and the announcement, as summarized, does not say how many led to arrests or what the false-positive rate was. That is exactly what a recurring report would need to show over time.
What changes in ChatGPT
Per Unite.AI, the changes fall into four buckets.
- Mass-attack content: for every user, regardless of age or stated reason, ChatGPT will give less detail on past attacks, including weapons, equipment and how they were obtained, and will avoid detailed comparisons of harm scale. Users can still ask about victims, community impact, warning signs and prevention.
- Warning signs: when recent chats show possible violent intent, ChatGPT responds more conservatively.
- Ending chats: ChatGPT can end a chat automatically if it shows fixation on violence. If this happens repeatedly, OpenAI may temporarily pause the account.
- Teens: if the system predicts a user is under 18, it applies stricter limits even when there is no sign of violent intent, including on practical information about weapons and security systems.
Teens can still get help de-escalating violence, find support and be pointed to a trusted adult, according to the same summary. OpenAI said the safeguards begin rolling out over the next few weeks.
How the under-18 prediction works
A lot rides on age prediction. Unite.AI says OpenAI strengthened its system, which uses behavioral and account signals to estimate whether someone is under 18. Users wrongly placed in the under-18 experience can verify adulthood through Persona. The rollout began January 20, 2026, and an August 25, 2026 update noted expansion to the EU.
The report also recaps earlier teen measures: ChatGPT for Teens launched August 18, 2026, applying default safeguards to users estimated to be under 18 or who state an age between 13 and 17. In July 2026 OpenAI began notifying parents when a teen's account is deactivated for violence-policy violations, and it is working on alerts for accounts paused for repeatedly discussing violence.
| Measure | Who it applies to | Stated timing |
|---|---|---|
| Less detail on past attacks | All users | Rolling out in the next few weeks |
| Stricter weapons and security-system limits | Predicted under-18 users | Rolling out in the next few weeks |
| Automatic chat ending, possible temporary pause | Users showing fixation on violence | Rolling out in the next few weeks |
| First Community Safety Report | Public | Early 2027, then every 6 months |
| Age verification via Persona | Adults wrongly flagged as under 18 | Available |
The trade-offs
Safety features that read intent are always a balance. Over-triggering can frustrate novelists, students, journalists and researchers who legitimately ask about history; under-triggering is the harm the company is trying to prevent. OpenAI's stated line is that discussion of victims, impact, warning signs and prevention stays open while operational detail narrows. Whether that holds in practice will only be visible after the rollout and the first report.
There are also privacy questions any referral program raises: what triggers a human review, what data is shared, and how disputes are handled. The announcement summary we saw does not answer those, and you should not assume specifics that were not stated.
If you use ChatGPT for research on sensitive history, ask about context, causes, victims and prevention rather than operational details. That is the area OpenAI says stays available. Adults misclassified as teens can verify through Persona.
01How many law-enforcement referrals did OpenAI report?
More than 200 in the third quarter of 2026, according to OpenAI's announcement as summarized by Unite.AI. It is a vendor-reported figure.
02When is the first full Community Safety Report?
Early 2027, covering the second half of 2026, with updates every six months.
03Do the new limits apply to adults?
The limits on past-attack detail apply to all users. The stricter weapons and security-system limits apply to users predicted to be under 18.
04Can ChatGPT end my conversation?
Yes. It can end a chat that shows fixation on violence, and repeated cases may lead to a temporary account pause.
05What if ChatGPT thinks I am a teenager?
Adults placed in the under-18 experience by mistake can verify their age through Persona.
Reading about a vendor's claims is one thing; testing them is another. With LumiChats you can put the same question to 40+ premium models side by side.
