Does ChatGPT report you to the police? When OpenAI does it, and how often

The Morse teamPublished November 4, 20268 min read

Yes, ChatGPT can report you to the police: as of November 4, 2026, OpenAI says that "when conversations indicate an imminent and credible risk of harm to others, we notify law enforcement" (OpenAI). OpenAI made "over 200 law enforcement referrals in Q3 2026," according to its October 30, 2026 post. OpenAI also reports child sexual abuse material to the National Center for Missing & Exploited Children (NCMEC), which passes the reports to police (OpenAI). An adult who talks about self-harm doesn't get referred, because OpenAI says it is "currently not referring self-harm cases to law enforcement" (OpenAI).

On November 4, 2026, we read OpenAI's safety posts from August 2025 to October 30, 2026, its February 2026 letter to Canada's minister of artificial intelligence, its usage policies, privacy policy, and government request policy, and its government request and child safety reports. OpenAI deserves credit for how much of this it has put in writing: its threshold, the reasoning behind a case it chose not to refer, and a count of its referrals.

"Reporting you" covers two different things. This page is about OpenAI going to the police on its own. Police can also come to OpenAI with a warrant, a subpoena, or an emergency request, which we cover in can AI chats be subpoenaed and AI transparency reports.

When does OpenAI report ChatGPT users to the police?

In three kinds of cases, by OpenAI's own account, and it publishes counts for two of them. Each row comes from OpenAI's own page, read November 4, 2026.

The teen row is a promise from September 2025, when OpenAI's parental controls post also said it was still "working on the right process" for reaching police when it can't reach a parent. None of the later posts we read says whether that process is now in place.

How does OpenAI decide to call the police?

Through a chain of automated flags and human reviews, set out in OpenAI's April 28, 2026 post on community safety:

1. Automated detection. OpenAI scans content and behavior with "classifiers, reasoning models, hash-matching technologies, blocklists, and other monitoring systems."
2. Human review. Trained staff look at a flagged chat in context, including "surrounding conversation, and any relevant patterns of behavior over time."
3. A ban, most of the time. Most enforcement, "including bans for violence," happens "directly between OpenAI and the user." A ban can cover "other accounts of the same user."
4. A deeper investigation, for a few. Cases with signs of "potentially serious, real-world harm" are assessed with "structured criteria," a stage OpenAI says is "reserved for a limited subset of cases."
5. A referral. Where the risk to others is "imminent and credible," OpenAI notifies law enforcement. Mental health and behavioral experts help with hard cases.

The bar doesn't require you to name a target. OpenAI calls its criteria "flexible," because a user "may not explicitly discuss the target, means, and timing of planned violence" and still pose a risk.

More automatic steps are coming. OpenAI's October 30 post says ChatGPT will be able to end a chat on its own when someone is "fixating on violence," and that "if ChatGPT ends multiple chats for that reason, we may temporarily pause the account." It says these safeguards begin rolling out "in the next few weeks."

How has OpenAI's police referral policy changed?

OpenAI's wording has gone from "may" to "we notify," and a mass shooting in Canada came in between. On February 10, 2026, a shooter killed eight people in Tumbler Ridge, British Columbia, and then herself, The Canadian Press reported.

OpenAI's February 26 letter to Evan Solomon, Canada's minister of artificial intelligence, says it banned a ChatGPT account of the shooter in June 2025, after automated detection and human review. It says reviewers "did not identify credible and imminent planning that met our threshold to refer the matter to law enforcement." The letter also says that under its newer protocol, "we would refer the account banned in June 2025 to law enforcement if it were discovered today." The shooter had opened a second account despite OpenAI's system for catching banned users, and OpenAI found it only after her name was made public.

British Columbia sued OpenAI in federal court in San Francisco in September 2026. The province alleges OpenAI's review team recommended a referral to the RCMP and that company leadership refused, CP24 reported. None of those allegations has been proven in court.

Here is how OpenAI's own wording has moved:

Many of the articles that rank for this question were written in 2025 and still quote only the "may refer" sentence.

Does ChatGPT report suicidal thoughts to the police?

OpenAI says it is "currently not referring self-harm cases to law enforcement" for adults. Teens in imminent danger are the one exception in the table above, and is it safe to tell ChatGPT personal things covers what happens to those chats instead. If you're thinking about suicide, you can call or text 988 in the US at any hour.

Will ChatGPT report you for asking about something illegal?

Not for a question alone, by OpenAI's account, since its pages name only violence against others and child sexual abuse as reasons it goes to police or NCMEC on its own. Its privacy policy, updated September 10, 2026, leaves wider room and lets OpenAI share data with "government authorities" to "detect or prevent fraud or other illegal activity."

Does OpenAI report to law enforcement when police ask?

Yes, with a warrant for chat content, a subpoena for account details, or an emergency request, which we cover in can AI chats be subpoenaed and count in AI transparency reports.

OpenAI now goes to police on its own more often than police come to it for chat content. In July to September 2026, OpenAI made more than 200 referrals for threats of violence. In the six months before, governments sent it 146 requests for chat content (OpenAI). Its CyberTipline reports to NCMEC ran to 126,597 in the same half-year, up from 75,027 a year earlier (OpenAI). OpenAI notes that the same user or the same image can be reported more than once. OpenAI's government request reports count only requests it received, so its referrals don't appear there. The first Community Safety Report, covering July to December 2026, is due in early 2027.

Will OpenAI tell you if it reports you?

None of the pages we read says it will. OpenAI's government request policy promises notice before it hands data to police who ask, "unless prohibited by law," and it may withhold notice in emergencies. That promise covers demands from a government, and no page we read extends it to referrals OpenAI starts itself.

You do get some warning in two places. If you've named a trusted contact, ChatGPT tells you when a chat is flagged for review, and the help page says that without that message "the conversation will not be sent for human review." A ban comes to you directly too, and OpenAI says that step is meant to make "clear they have crossed a line."

How Morse keeps your everyday chats off a scanned server

Every step above depends on ChatGPT keeping your conversations on OpenAI's servers, where classifiers can sweep them and reviewers can read "patterns of behavior over time" across them. Morse is built without that archive. Your chat history is stored only in your browser, encrypted, and our servers don't keep your messages or log what you write. Because Morse doesn't store chats, it usually has very little to give authorities. Like OpenAI, Morse reports child sexual abuse material to NCMEC.

For the grief, the anger, or the dark short story you'd rather not leave in a monitoring queue, choose a mode for each chat:

• E2EE, for the writing you'd want no reviewer to read. Before a message leaves your device, your browser checks the secure enclave's hardware attestation, then encrypts the message so only that enclave can open it. Our servers relay ciphertext they can't read, and the reply is decrypted only in your browser. The enclaves are run by third-party providers, not by Morse, and the guarantee rests on the enclave hardware and attestation working as designed. Learn how end-to-end encrypted AI chat works.
• Private, the default, for chats you'd rather no host kept. Open-weight models like DeepSeek, Kimi, and GLM, plus xAI's Grok, answer this mode's chats. Morse has zero data retention agreements with its inference, search, and other providers, so they keep nothing. Where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't.
• Anonymous, for GPT, Claude, and Gemini without the provider knowing it's you. Our proxy removes your account, email, and IP address and sends the message under Morse's own account. The provider still sees the message and may keep it for a limited time to check for abuse, under its own policy. See how to use ChatGPT anonymously.

Your history locks when you sign out, and opening it takes two keys: one that never leaves your device, and one our server gives you at sign-in. Clearing your browser deletes your chats for good, and Morse can't recover them, so keep a copy of anything you need. In character chats, Morse shows crisis lines when a message reads like suicidal thinking, and that check runs on your device.

Start a chat from the Morse home page without an account, or sign up free with no card. End-to-end encrypted chats come with Pro, at $19 a month.