People now treat AI chatbots as diary, therapist, and lawyer at once — typing out their plans, their fears, and their worst impulses in full sentences. None of it is privileged. All of it is stored. And starting last fall, the government began asking for it. The Supreme Court's June decision in Chatrie v. United States, 609 U.S. ___ (2026), never mentions artificial intelligence. Yet it undercuts the argument the government leans on hardest — that by typing into a company's system, a user has already surrendered any privacy in what he wrote.
The warrants are already here. In June 2026, a Manhattan federal judge refused to quash a warrant to OpenAI for a securities-fraud defendant's ChatGPT records, declining to reach his argument that the search would sweep in privileged defense material, though the court left him free to assert privilege over specific items later. United States v. Kim, No. 1:25-cr-00359-LGS (S.D.N.Y.). Months earlier, in a separate Manhattan case, a federal judge ruled that a defendant's conversations with Anthropic's Claude were not protected by attorney-client privilege. And the first known federal warrant to OpenAI, unsealed in Maine in the fall of 2025, sought a user's ChatGPT history in a dark-web investigation. Whatever the client typed, the government can now try to read.
What Chatrie decided, and the principle that travels
The mechanics of Chatrie are narrow. Police used a geofence warrant to place a suspect at a credit-union robbery. The Court held, 6-3, that acquiring that location data was a Fourth Amendment search and sent the case back to decide whether the warrant was reasonable.
Two moves in Justice Kagan's opinion are what may impact AI.
Second, the Court described the protected data as the user's own. Location history, it said, functions as "a personal journal" that the user consults and edits, and in that way "resembles other private materials — think of emails, documents, photographs, or calendars — that even if stored on Google's servers, a user reasonably views as his own." That sentence is the bridge to this whole problem. An AI chat log is the purest example of the thing the Court described: a running, first-person record of a person's thinking, held on a company's servers, that the user treats as private.
A word on why this matters. Courts have leaned for years on a comfortable fiction, suggesting that people choose to surrender their privacy when they hand data to a company. For most of modern life, there is no choice. The technologies that participation now requires are built to extract disclosure as the price of entry, and calling that "voluntary" describes the law's convenience, not the user's experience. Chatrie is the Court saying so, and it is overdue.
So the principle Chatrie stands for — that handing data to a company to use its service is not a waiver, and that a personal record does not lose protection because a provider stores it — maps directly onto the prompts your client typed into ChatGPT or Claude. The reasoning travels even though the holding does not.
Why Chatrie alone won't protect a chat log
Here is the discipline the analysis requires, and the place a careless brief will fail. Chatrie is a location case. It rests on the privacy interest in "the whole of [a person's] physical movements" that Carpenter v. United States, 585 U.S. 296 (2018), recognized. A chat log is not location data. So Chatrie supplies the principle and the rhetoric, but it does not, by itself, establish a Fourth Amendment interest in the content of a prompt.
The authority that does that is the stored-communications line. In United States v. Warshak, 631 F.3d 266 (6th Cir. 2010), the Sixth Circuit held that a person keeps a reasonable expectation of privacy in the content of emails held by a commercial provider, that the government must get a warrant to compel them, and that the Stored Communications Act is unconstitutional to the extent it allows warrantless access to content. Warshak is the chassis; Chatrie is the engine. A prompt is content in exactly the sense Warshak protected — a communication the user composed and entrusted to a provider — and the government's "he shared it with the company" argument runs into both Warshak's holding and Chatrie's refusal to treat that kind of sharing as a waiver.
A prompt is more revealing than a search
A Google search is a few words. A prompt is often a paragraph — a person reasoning aloud, disclosing motive, sequence, and state of mind in a way a search box never captured. Investigators know it. The evidentiary value of these logs is precisely that people write to a chatbot the way they would to a confidant.
That expressive quality is itself a legal hook. In People v. Seymour, 536 P.3d 1260 (Colo. 2023), the leading reverse-keyword-warrant case, the Colorado Supreme Court held that a person has a privacy interest in Google search history and that, because a search implicates free expression, a warrant seeking it must be evaluated with "scrupulous exactitude." The court grounded the privacy holding in the Colorado Constitution, noting that federal courts often reject the same expectation under the third-party doctrine, so a Florida practitioner cannot import it wholesale. But the expression point is federal, and it applies with more force to a prompt than to a keyword: the stronger challenge to a demand for chatbot content may sound in the First Amendment as much as the Fourth.
Distinguish two scenarios, because they are litigated differently. The demands seen so far — Kim, the Claude case, the Maine warrant — are targeted: the government identifies a suspect and seeks his chats. Those are content fights under Warshak, won or lost on probable cause, particularity, and overbreadth. The other scenario is the reverse demand — "identify every user who asked the chatbot about X" — the dragnet analog to a geofence. That version runs straight into the general-warrant bar that sank the geofence warrants in United States v. Smith, 110 F.4th 817 (5th Cir. 2024), and it is the scenario civil-liberties groups warned the Court about in Chatrie. Know which one you are fighting.
The two traps: retention and privilege
Counsel your clients on both, because both defeat the assumptions ordinary users make.
The first is retention. Users believe "delete" means gone. It does not. In the copyright litigation brought by The New York Times, a federal court ordered OpenAI to preserve output logs — including chats users had deleted and sessions they marked "temporary." That going-forward order was later scaled back, but data already preserved stayed producible. Retention varies by provider, by consumer-versus-enterprise tier, by account settings, and by jurisdiction, but the safe assumption is that the data persists and is producible. A client who "cleared" a conversation has cleared his screen, not the server.
The second is privilege. This is where the white-collar exposure is sharpest. In Kim, the defendant said he had used AI tools for "case-related research" and argued the warrant would expose his defense strategy; the government answered that chatbot exchanges are not privileged because a chatbot is not a lawyer, gives no legal advice, and supports no reasonable expectation of confidentiality. The court in the Claude case reached the same conclusion. A client who pastes the facts of his case into a chatbot to "figure out his options" is not consulting counsel — he is creating a discoverable record, and if he pastes in genuinely privileged material, he may be waiving the protection that attached to it. Clients need to hear this before they are charged, not after.
What to do now
The threshold argument is strong and getting stronger: acquiring the content of a person's chat log is a search requiring a warrant, and the "he shared it with the company" answer fails under Warshak and Chatrie alike.
But this is an unsettled area of law, and astute practitioners need to ensure they are making a careful record in the trial court while the doctrine is open. This area has no controlling Eleventh Circuit or Supreme Court answer to the prompt-content question. The facts you develop are what an appellate court will need when the answer comes.
The takeaway
Chatrie is not an AI case, and its real subject is already obsolete. Its value is the principle it restates at the highest level: giving your data to a company to use a service is not consent to give it to the government, and a personal record stays yours even on someone else's servers. That principle, joined to Warshak's protection of stored content and Seymour's recognition that expression demands exacting scrutiny, is the framework for the fight now arriving over the chat logs our clients created believing no one else would ever read them.
