Can AI chats be subpoenaed? Yes, and here's how it has already happened

The Morse teamPublished October 15, 20269 min read

Yes. AI chats can be subpoenaed or ordered produced in a lawsuit, seized under a search warrant, or ordered preserved by a court, and in 2025 and 2026 all three happened. They aren't protected by any special "AI privilege": a federal judge in New York ruled in February 2026 that a defendant's chats with Claude were not covered by attorney-client privilege.

This is general information, not legal advice. If you're in a legal dispute or expect one, talk to a lawyer before you type anything about it into a chatbot.

Are AI chats privileged?

No, not on their own. The clearest ruling so far is United States v. Heppner, a securities fraud case in the Southern District of New York. Bradley Heppner had used Claude, by himself, to write up defense strategy after he received a grand jury subpoena. When he was arrested in November 2025, FBI agents executed a search warrant at his home and seized "approximately thirty-one documents that memorialize communications" with Claude, according to Judge Jed S. Rakoff's memorandum, filed February 17, 2026.

The judge ruled from the bench on February 10 that the documents were protected by neither attorney-client privilege nor the work product doctrine. (Some summaries give February 13; the docket shows the bench ruling on February 10 and the written memorandum on February 17.) His reasons, in plain words:

• Claude isn't a lawyer. "Because Claude is not an attorney, that alone disposes of Heppner's claim of privilege."
• The chats weren't confidential. The privacy policy Claude users agreed to said Anthropic collects inputs and outputs, uses them to train Claude, and may disclose data to "third parties," including "governmental regulatory authorities."
• Sending them to a lawyer later didn't help. Non-privileged communications "are not somehow alchemically changed into privileged ones upon being shared with counsel."
• No lawyer directed it. His own counsel conceded they "did not direct" him to use Claude, so the chats weren't work product either.

The judge left one door open. Had counsel directed Heppner to use Claude, he wrote, Claude "might arguably" have acted like a lawyer's agent inside the privilege.

Not every court sees AI the same way. On the same day as the Heppner bench ruling, a magistrate judge in Michigan held in Warner v. Gilbarco that a person suing without a lawyer could protect her ChatGPT use as work product, calling AI programs "tools, not persons," according to an analysis by Wilson Elser. The law is unsettled, so nobody should count on privilege for a chat with a consumer chatbot.

OpenAI's chief executive said as much in July 2025. On Theo Von's podcast, Sam Altman said that with a therapist, a lawyer, or a doctor "there's legal privilege for it," and "we haven't figured that out yet for when you talk to ChatGPT." He called that "very screwed up."

How have AI chats ended up in court?

Many articles on this question still say OpenAI must keep every ChatGPT chat "indefinitely." That stopped being true in September 2025. Here is the record, from the court filings and the companies' own reports.

The 20 million logs were ordinary users' conversations, ordered handed to the other side in a copyright case those users had no part in. The judge's reasoning also rests on the fact that OpenAI, like most AI services, keeps the chats "in the normal course of its business."

Who can get your AI chats, and how?

There are three routes, and they work differently.

In a lawsuit: usually from you

In civil cases (a divorce, an employment dispute, a contract fight), the other side asks for relevant records in discovery. Whether it can subpoena the AI company directly is unclear. An analysis in the bar journal Colorado Lawyer (May/June 2026) argues that the federal Stored Communications Act "does not create an exception for civil discovery," so courts are likely to find providers can't be made to hand over chat content on a civil subpoena. It also notes that no reported decision has settled the point.

Instead, the request goes to you. If you can log in and export your chat history, the same analysis says, "courts are likely to find these materials within the producing party's control," so a court can order you to produce them. And when the AI company is itself a party, as OpenAI is in the Times case, it can be ordered to produce its own logs, as the 20 million-log order shows.

In a criminal case: a warrant to the company

OpenAI's policy for government requests, effective January 1, 2026, says it "only discloses requested user content to a law enforcement request in response to a valid warrant or equivalent," and needs "a subpoena, court order, search warrant or equivalent" for account details. On a valid request it will also preserve an account's data for 90 days, and another 90 on request, while police get the paperwork. Its reports show how often that happens:

Sources: OpenAI's reports for the second half of 2025 and the first half of 2026, and Anthropic's report for July to December 2025, its latest listed on its transparency page as of October 15, 2026. National security requests are reported separately, in ranges.

Content requests to OpenAI nearly doubled in six months, from 75 to 146, though the numbers are still small.

Your own device

Heppner took this route. The FBI seized documents and devices from his home instead of subpoenaing Anthropic. Whatever is on your laptop or phone, including chat history, exports, and screenshots, is reachable by a warrant to search that device. Keeping history off a company's servers changes who can be asked for it, but the copy on your device can still be seized.

Does deleting your AI chats help?

Less than you'd think. OpenAI says a deleted chat is scheduled for permanent deletion "within 30 days," unless it was already de-identified or OpenAI "must retain it longer for security or legal obligations" (OpenAI). From May to September 2025, a court order made OpenAI keep the ChatGPT output logs it would otherwise have deleted.

Timing matters more. Once a lawsuit is reasonably expected, deleting relevant records can be held against you. Under Federal Rule of Civil Procedure 37(e), if a court finds a party destroyed electronic information "with the intent to deprive" the other side of it, it may tell the jury to presume the information was unfavorable, or dismiss the case. If you want to keep less, do it before there's a dispute.

What reduces your exposure?

These steps can't put a chat out of reach, but they leave fewer copies, in fewer places, for a shorter time.

1. Don't work through your own legal problem in a chatbot. Heppner turned on that exact habit. Take the questions to your lawyer, and if you want to use AI, ask them which tool and how.
2. Know where your history lives. A chat stored on a company's servers can be preserved, sampled, or produced by that company. Our comparison of what ChatGPT, Claude, Gemini, Grok, and Meta AI keep shows how long each holds chats.
3. Use temporary chats for sensitive one-offs. They're kept for a shorter time, though OpenAI still keeps them for up to 30 days.
4. Keep fewer copies yourself. Exports, screenshots, and synced notes are records too.
5. Lock your devices. A device search is the route that worked in Heppner.
6. Don't delete once a dispute is on the horizon. Ask your lawyer what to keep.

For a wider look at who can see your chats beyond the courts, from employers to human reviewers, see how to use AI privately.

How Morse keeps your chats off company servers

Judge Stein's order rested on OpenAI keeping the chats "in the normal course of its business." A court can order a service to produce what it stores, so Morse is built not to store your chats. Your chat history is kept only on your device, in your browser, encrypted and locked when you sign out. Our servers don't keep messages or log message content. Your account holds your sign-in details, plan, balance, and a usage ledger with no content. So when the law requires Morse to hand something over, our privacy policy says it plainly: "Because we don't store your chats, we usually have very little to give."

Each chat also has a privacy mode, so you decide who handles each message:

• Anonymous, to use GPT, Claude, and Gemini without the provider knowing it's you. Our proxy removes your account, email, and IP address before the message goes out, and sends it under Morse's own account. The provider sees the question, with your identity stripped.
• Private, for questions you'd rather no host keep. It's the default for a new chat. Open-weight models like DeepSeek, Kimi, and GLM, plus Grok, run on hosts that have agreed to zero data retention: they keep neither prompts nor replies. Where a host confirms that on each reply, Morse checks it and refuses a reply that doesn't.
• E2EE, for anything no one should read. Your browser checks the secure enclave's proof that it's genuine before anything leaves your device, then encrypts each message so only that verified enclave can open it. Our servers relay text they can't read, and the reply is opened only in your browser. Read how end-to-end encrypted AI chat works.

Your device is where your history lives, so it's the one place to protect. 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. Sign out or lock your device when you step away. For a question you don't want saved anywhere, start a Temporary chat: it's never saved, not even on your device. And because there's no server copy, what you clear from your browser is gone for good.

Ask your next sensitive question on Morse. Try a chat from the Morse home page without an account, or start free with no card needed. End-to-end encrypted chats come with Pro, at $19 a month.