Do AI companies notify you of legal requests? What 12 of them promise

The Morse teamPublished November 12, 202610 min read

OpenAI does notify users of subpoenas and police demands before it hands over their data, unless a court or the law forbids it, and since April 2026 it has promised notice for subpoenas in civil lawsuits too, "where legally possible and appropriate." Of the 12 AI companies we checked, six promise in writing to tell you first: OpenAI, Anthropic, Google, Microsoft, Meta, and Apple. Five have no notice promise we could find: xAI, Perplexity, Character.AI, Mistral, and DeepSeek. Proton, which makes Lumo, says Swiss law leaves the telling to the authorities.

We read each company's own law enforcement guidelines, legal process pages, and privacy policies on November 12, 2026, and quote them below. Every promise comes with exceptions, and the biggest is out of the company's hands, because a court can order it to keep quiet. This isn't legal advice.

Which AI companies notify you of legal requests?

Six of the twelve, before they disclose anything, as of November 12, 2026. "Content" means your prompts and the replies. In the US, a gag order is usually a court order under 18 U.S.C. § 2705(b), which we explain further down.

The four big platforms don't name their assistants on these pages. Google's policy, Meta's guidelines, Microsoft's report page, and Apple's guidelines don't mention Gemini, Meta AI, Copilot, Siri, or Apple Intelligence, so we read each promise as covering the whole account.

Will ChatGPT tell me if police ask for my data?

Yes, in most cases, and before the data goes out. OpenAI's Government User Data Request Policy, effective January 1, 2026, says that when it receives "a demand from a government to disclose information about a user, OpenAI provides notice to the user before disclosing the information, unless prohibited by law."

It may skip the notice in two kinds of case: "emergencies where notice could result in danger," and where notice "would be counterproductive (e.g., where an account has been compromised)." When a gag order runs out, OpenAI "provides delayed notice to users after disclosure." The policy also covers requests sent to OpenAI Ireland, which serves users in the European Economic Area and Switzerland.

That promise covers police coming to OpenAI. OpenAI going to police on its own, when it sees a threat of violence in a chat, is a different process, and no OpenAI page we read promises notice for it. We cover that in does ChatGPT report you to the police.

Does OpenAI notify users of civil subpoenas?

Yes, since April 24, 2026. OpenAI's government policy still says it "is not intended for requests for information or service of legal process by OpenAI users or other civil or criminal litigants." On its own, that line reads as if OpenAI had no civil policy. But a separate page for civil requests, published April 24, 2026, says: "We will notify users before any data is disclosed, where legally possible and appropriate."

"Where legally possible and appropriate" is looser than the government policy's wording, and the page doesn't define it or promise delayed notice.

The others that cover civil cases:

• Microsoft says it "will give prior notice to customers whose data is sought by a civil proceeding litigant," and that for content requests "we require specific lawful consent of the account owner."
• Apple extends its notice to requests from "third parties," not just governments.
• Meta sends "requests from civil litigants and criminal defendants" to its Facebook and Instagram Help Centers, outside its law enforcement guidelines.
• xAI says legal requests "from non-government entities (including private entities) must be served on" its registered agent, and says nothing about telling you.

Whether a litigant can get chat content from the company at all is a separate question, which we cover in can AI chats be subpoenaed.

When can an AI company stay silent?

When a court orders it to, and in a few cases the company decides for itself.

Court orders are the usual reason. Under 18 U.S.C. § 2705(b), a court "shall" order a provider not to tell anyone about a warrant, subpoena, or court order if there's reason to believe notice would endanger someone's life or safety, lead to flight from prosecution, destruction of evidence, or intimidation of witnesses, or "otherwise seriously jeopardiz[e] an investigation." Microsoft, which publishes the figure, says that in the second half of 2025 secrecy orders were attached to 32% of the US legal demands it received: 1,823 orders, 1,473 of them from federal authorities. Our AI transparency reports page has Meta's share, which is higher.

The company's own exceptions follow a pattern: emergencies, child safety, and hijacked accounts. They differ in what happens next:

• Google says that after an emergency, "we'll provide notice if we learn that the emergency has passed."
• Apple sets a clock: "In emergency disclosure cases, Apple will provide delayed notice after the expiration of 90 days."
• OpenAI, Microsoft, and Meta promise delayed notice when a nondisclosure order expires.
• Anthropic's page makes no promise of later notice.

Microsoft says it has challenged unnecessary secrecy orders "both directly in communications with law enforcement and formally in court," and has asked Congress to narrow § 2705 so that such orders are "time-limited."

Do police need a warrant to get your AI chats?

Yes, at five of the twelve, by their own US policies: OpenAI, Google, Microsoft, Meta, and Apple. OpenAI "only discloses requested user content to a law enforcement request in response to a valid warrant or equivalent." Apple requires "a search warrant issued upon a showing of probable cause, or customer consent." A subpoena gets less: Google and Meta list basic subscriber details such as your name and email address, and some IP addresses.

Anthropic and xAI don't say. Anthropic's policy page, dated March 16, 2026, requires "valid legal process (eg., a validly issued subpoena or warrant)" without saying which one reaches your prompts. Its July to December 2025 report says only that "a search warrant may request a user's content." xAI's page, which now names the company SpaceX Super Intelligence LLC, lists "a subpoena, court order, or warrant" together.

All six companies that promise notice also allow emergency disclosure without a warrant. Google gives "bomb threats, school shootings, kidnappings, suicide prevention, and missing persons cases" as examples, and Apple says that in an emergency federal law leaves it "permitted, but not required," to disclose.

Encryption changes what a warrant can reach. Meta's guidelines say that for chats protected by end-to-end encryption, "Meta will no longer be able to provide unencrypted message content as part of any response to any legal process."

Which AI companies don't promise to tell you?

Five, as of November 12, 2026. Each says it shares data with authorities and says nothing about telling you:

• xAI. Its law enforcement page says non-public information "will not be released except as required by appropriate legal process." Its privacy policy, last updated October 27, 2026, allows sharing "to respond to lawful requests and legal process."
• Perplexity. Its privacy notice, last updated October 23, 2026, says it "may share data to cooperate with law enforcement, participate in a legal process, or for legal compliance," and offers regulators and police a contact form.
• Character.AI. Its privacy policy, effective July 1, 2026, may disclose information "to respond to a warrant, subpoena, court order, or other applicable laws and regulations."
• Mistral. Its privacy policy, effective September 3, 2026, lists "competent courts" among those it may share data with, and has no law enforcement section. Its data processing addendum for business customers does promise to "promptly inform Customer" of a legal requirement to process their data, "unless prohibited to do so by applicable law and/or on important grounds of public interest."
• DeepSeek. Its privacy policy, last updated February 10, 2026, shares data to comply with "legal process or government requests" and says "we may provide Personal Data (such as your IP address) to law enforcement agencies in the event of an emergency."

Proton is a different case. Its privacy policy says it discloses only on "a binding request coming from the competent Swiss authorities," that its "general policy is to challenge requests whenever possible" where it doubts their validity or sees a public interest, and that it can't "decrypt end-to-end encrypted content." Lumo's privacy policy, last modified September 17, 2026, says saved chats are stored with "zero-access encryption (meaning we as a provider are not able to decrypt it anymore)." Proton deserves credit for this. By its own account, a request that reaches it can't turn saved Lumo chats into readable text. Our Proton Lumo review covers what Lumo does see while it answers.

How Morse keeps your chat history off the server a request is sent to

Every promise in the table is about a copy the company already has. A notice email gives you time to call a lawyer, but it doesn't change what the company must produce, and a gag order can hold the email back until the order runs out. Morse is built so that copy isn't on our servers. Your chat history is stored in your browser, encrypted. Our servers don't keep your messages or log what they say, and your account holds only your sign-in details, plan, balance, and a usage ledger with no content. Because Morse doesn't store chats, it usually has very little to give authorities.

You also pick, per chat, who handles each message on its way to the model:

• Private, for chats you'd rather no AI host kept. It's the default for a new chat, with open-weight models like DeepSeek, Kimi, and GLM, plus xAI's Grok. 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.
• E2EE, for anything no one should be able to read. Before a message leaves your device, your browser checks the secure enclave's proof that it's genuine, then encrypts the message so only that enclave can open it. Our servers relay ciphertext they can't read. The enclaves are run by third-party providers, not by Morse, and the guarantee rests on the enclave hardware and attestation working as designed.
• Anonymous, for GPT, Claude, and Gemini without the provider knowing it's you. Our proxy strips 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, so leave out names and details you'd rather keep to yourself.

With the history on your device, the device is what to protect. Opening it takes two keys, one that never leaves your device and one our server hands you at sign-in, so a copy of your browser data alone can't be read. Lock your screen when you step away. Clearing your browser deletes your chats for good, and Morse can't recover them, so save anything you need. Morse doesn't train AI on your chats, sell your data, or show ads.

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.