What is a warrant canary? What it means, whether it's legal, and who still has one

The Morse teamPublished November 7, 20268 min read

A warrant canary is a statement a company publishes, and keeps renewing, saying it has not received a secret government demand for user data, so that if the statement goes stale or disappears, readers can infer that a demand has arrived. The Electronic Frontier Foundation (EFF) calls it "a regularly published statement that a service provider has not received legal process" that a gag order would stop it from mentioning, and the name refers to "the canaries used to provide warnings in coalmines" (EFF).

We read EFF's canary FAQ, the US statutes behind the gag orders, two live canaries, Cloudflare's transparency report, Apple's 2013 and 2014 reports, and the privacy and transparency pages of nine AI chat services, all on November 7, 2026. Of the AI chat services we checked, only Kagi publishes a canary.

Who publishes a warrant canary in 2026?

Read November 7, 2026, on each company's own page.

Why is it called a warrant canary?

Because, like the caged birds miners once carried underground, it warns by going silent. EFF's FAQ draws the coal mine comparison itself (EFF).

A plain version is a printable sign for libraries on librarian.net, the site of librarian Jessamyn West. It reads "the FBI has not been here. watch closely for removal of this sign," with a note that it rests "on the assumption that it's only illegal to say they've been there if it's true" (librarian.net).

rsync.net, a cloud storage company, still publishes one. Its notice promises a weekly "warrant canary" listing "all warrants and subpoenas served since 2006," and it asks readers to take "special note" if the messages "ever cease being updated, or are removed from this page" (rsync.net). The message dated November 2, 2026 lists "(none)" for each of its four locations, and carries Reuters headlines and two basketball scores from November 1 to prove it wasn't written in advance.

Is a warrant canary legal?

In the US, probably, but no court has ruled on one. Asked whether courts have upheld canaries, EFF's FAQ answers "Not yet" (EFF). This isn't legal advice.

The gag that a canary works around is written into federal law. Under 18 U.S.C. 2709, a provider that receives a National Security Letter may be barred from telling "any person that the Federal Bureau of Investigation has sought or obtained access" to the records (18 U.S.C. 2709). Orders from the Foreign Intelligence Surveillance Court and ordinary subpoenas can carry gags of their own, as EFF's FAQ lists.

The case for legality rests on the difference between silence and a lie. EFF writes that the government "may be able to compel silence through a gag order," but "it may not be able to compel an ISP to lie by falsely stating that it has not received legal process" (EFF). Its advice to a provider that does receive a gagged demand is to "obtain legal counsel and go to a court."

The harder question is whether a gag could order a company to keep its canary up. The ACLU calls a canary a "legally experimental" statement, and says whether such a demand would violate the First Amendment "is a thorny question" (ACLU).

Do warrant canaries work?

As a signal, sometimes. Apple's and Reddit's canaries both went quiet, and in each case readers could see the line was gone but not what had arrived.

Apple's first transparency report, dated November 5, 2013, ended with "Apple has never received an order under Section 215 of the USA Patriot Act. We would expect to challenge such an order if served on us" (Apple). Its report for January to June 2014 has no such sentence. In its place, it reports national security orders in "bands of 250," with "0 - 249" received (Apple).

Reddit's 2014 transparency report said that "as of January 29, 2015, reddit has never received a National Security Letter, an order under the Foreign Intelligence Surveillance Act, or any other classified request for user information." The ACLU noted on April 4, 2016 that the report for 2015, issued the week before, dropped the line (ACLU).

Canaries also go quiet for dull reasons. EFF's look back at its Canary Watch project, on May 25, 2016, found that "canaries often were not updated at all, or were updated several days or weeks late," and that some would "disappear and come back later, unchanged" (EFF). rsync.net says of its own process that it "is not infallible," and that its signature stops others from forging a message but not someone from coercing rsync.net into publishing a false one (rsync.net).

Why a canary says "0" and a transparency report can't

US law lets companies report national security requests only in ranges, and every range starts at zero. The statute's bands are "0–999," "0–499," "0–249," and, for yearly totals, "0–99" (50 U.S.C. 1874). A company's first order on a new platform can be left out of reports "until 540 days after" it arrives.

Google's report shows what a band hides. For July to December 2025, it lists "0 – 499" FISA requests for content, affecting "211000 – 211499" accounts (Google). The request band includes zero, and the account count makes plain that the true number of requests wasn't zero.

A canary is the one place a company can say the number is exactly zero. EFF's FAQ makes the same point, that canaries can be more specific than the government's ranges. Swapping a canary for a band is itself a signal. EFF's Canary Watch saw Pinterest change "National Security: 0" to a band of 0-249 (EFF), the same move Apple's 2014 report made.

Which AI companies have a warrant canary?

As of November 7, 2026, Kagi is the only AI chat service we checked that publishes one. It sits in Kagi's privacy policy and says Kagi has received "0 National Security letters," "0 Gag orders," and "0 Warrants from any government organization" (Kagi). Kagi deserves credit for keeping it in its main policy. The canary itself carries no date, though, so a reader can't tell when it was last confirmed, only whether it's still there. We cover how Kagi keeps Assistant threads in is Kagi Assistant private.

Venice users have asked for one. A request titled "Warrant Canary" on Venice's feedback board, posted February 14, 2025, had 66 votes and the status Backlog on November 7, 2026 (Venice feedback). Venice's privacy policy says it won't "have access to or store the content of your Prompts or Outputs," apart from generated videos kept until you download them (Venice).

The largest AI companies publish transparency reports instead. OpenAI's report for January to June 2026 says it received "250-499" national security requests, "within the ranges permitted by law" (OpenAI), and Anthropic's Transparency Hub links reports through July to December 2025 (Anthropic). We set their counts side by side in AI transparency reports.

Proton publishes a transparency report with no canary. It says Proton "may not transmit any data to foreign authorities directly," and counts orders for Mail and VPN but not for Lumo (Proton). Confer answers the question a canary is about without one: asked "What if someone subpoenas my conversations?", its home page says "They'd get ciphertext. So would we" (Confer).

What protects your chats when a canary goes quiet?

A canary tells you a demand arrived. What the demand gets depends on what the company holds. In April 2023, Swedish police came to Mullvad VPN's office in Gothenburg intending "to seize computers with customer data." Mullvad says "such customer data did not exist," and the police "left without taking anything and without any customer information" (Mullvad).

Morse is built on the same idea for AI chats. Your chat history is stored in your browser, encrypted, and our servers don't keep messages or log what you write. Your account holds 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. Each chat has its own privacy mode:

• E2EE, for anything no one should read. Your browser checks the secure enclave's hardware attestation, then encrypts the message so only that enclave can read it, and 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. E2EE runs open-weight models like GLM, Kimi, and DeepSeek, without web search or image generation, and comes with Pro, $19 a month.
• Private, so there's no copy at a model host for a request to reach. It's the default for a new chat. 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, so a request to OpenAI, Anthropic, or Google doesn't lead back to your name. Our proxy removes your account, email, and IP address and sends the message to GPT, Claude, or Gemini under Morse's own account. The provider sees the message without knowing it's you, 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.

Your history lives on your device, so that's the copy to guard. Unlocking it takes two keys, one that never leaves your device and one our server gives you when you sign in, so a copy of your browser data alone can't be read. Clearing your browser or deleting your account deletes your chats for good, and we can't recover them. Morse doesn't train AI on your chats, sell your data, or show ads. Read the details in our privacy policy. For how other apps store chats, see where AI chat data is stored, and for how chats reach a courtroom, can AI chats be subpoenaed.

Start a chat on the Morse home page without an account, or sign up free with no card and add E2EE with Pro.