MorseSign inStart freeOpen Morse

Does the Fourth Amendment protect your ChatGPT conversations? What US law says

The Morse teamPublished 13 min read

No court has decided yet whether the Fourth Amendment protects ChatGPT conversations, but as of November 2026 the case law suggests it does for the chats themselves. In June 2026 the Supreme Court held that police conducted a search when they got a user's Location History from Google, and likened it to "emails, documents, photographs, or calendars" that a user "reasonably views as his own" even when Google stores them (Chatrie). A federal appeals court held in 2010 that the government needs a warrant to make a provider hand over the contents of email (Warshak). The account details behind a chat, such as your name, address, and IP addresses, get much less protection, because a federal statute lets the government get them with a subpoena (18 U.S.C. § 2703).

On November 19, 2026, we read the Supreme Court's opinions in Miller, Smith, Riley, Carpenter, and Chatrie, the Sixth Circuit's Warshak opinion, the Colorado and Pennsylvania supreme courts' decisions on Google search records, the Stored Communications Act, the warrant papers in the known cases involving OpenAI, xAI, and Anthropic, and a May 2026 Congressional Research Service analysis. We looked at US law only. This isn't legal advice.

The companies already set a higher bar than the statute's text in places. 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." We compare 12 companies' policies in do AI companies notify you of legal requests. If a policy like that changed, the law is what would be left.

What is the third-party doctrine, and does it cover AI chats?

It's the rule that you give up Fourth Amendment protection in what you hand to a company, and it was built on bank records and dialed phone numbers, not on what you say. In United States v. Miller (April 21, 1976), the Supreme Court held that a bank customer had no Fourth Amendment interest in checks and deposit slips the bank kept, because "the depositor takes the risk, in revealing his affairs to another, that the information will be conveyed by that person to the Government" (opinion).

Smith v. Maryland (June 20, 1979) applied the same idea to the numbers a person dials. "A person has no legitimate expectation of privacy in information he voluntarily turns over to third parties," the Court wrote (opinion). But the Court stressed that a pen register differs from a wiretap because pen registers "do not acquire the contents of communications."

A prompt to ChatGPT is content, the words you typed and the reply, so the doctrine's reach over AI chats is still open. A Congressional Research Service sidebar of May 5, 2026 said: "It appears that federal courts have not yet had the occasion to determine whether a user maintains a reasonable expectation of privacy in chatbot histories, or whether that expectation is forfeited by virtue of the third-party doctrine."

The Fourth Amendment limits only the government. When OpenAI's own systems review your chats, its terms govern that. The amendment applies when police compel the company to search or hand over your data, which Justice Gorsuch's Chatrie opinion describes as a search "effected" by a private party acting "as an instrument or agent of the Government."

Did Carpenter change the rule for data tech companies store?

Yes, for one kind of data, and the Court said so narrowly. In Carpenter v. United States (June 22, 2018), police had used court orders to get a suspect's cell-site records from his wireless carriers, 12,898 location points over 127 days (opinion). The Court held that getting those records was a search, even though a third party held them.

It also rejected the statutory order police had used. A Stored Communications Act order under section 2703(d) requires only "reasonable grounds" to believe records are "relevant and material" to an investigation, which "falls well short of the probable cause required for a warrant." So before compelling a carrier to hand over location records, the Court wrote, the government's obligation is "a familiar one": "get a warrant" (Carpenter).

Then it fenced the ruling in: "Our decision today is a narrow one." It left out real-time tracking and "tower dumps," and said, "We do not disturb the application of Smith and Miller" (Carpenter). That left lower courts free to treat other records held by companies under the old rule, and the Colorado Supreme Court noted in 2023 that "most federal courts continue to reject" a privacy interest in information shared with a third-party vendor.

What did the Supreme Court decide in Chatrie?

That a person "has a reasonable expectation of privacy in records about his cell phone's location," even when only a short stretch is taken, and "from a third-party tech company." Chatrie v. United States was decided June 29, 2026, by 6 to 3 (opinion). Justice Kagan wrote for five justices, and Justice Gorsuch agreed with the result.

The opinion never mentions AI chats, but three of its points reach them:

  • Stored data can still be yours. Location History resembles "other private materials," such as "emails, documents, photographs, or calendars," that "even if stored on Google's servers, a user reasonably views as his own."
  • Opting in doesn't give it away. The government argued that turning on Location History was a voluntary choice. The Court said that argument "ignores some pertinent facts" about how people come to turn it on, and that "a cell-phone user is not to be viewed as sharing private information with third parties" by "doing the ordinary things cell-phone users do."
  • It didn't decide whether the warrant was valid. The Court sent the case back to the Fourth Circuit to decide whether each step of the geofence warrant met the Fourth Amendment's demands for "particularity and probable cause."

Justice Gorsuch, who agreed with the result on his own reasoning, wrote that "entrusting your effects to a third party for certain agreed purposes doesn't mean they are no longer yours" (Chatrie).

Does the Fourth Amendment require a warrant for stored chat contents?

For email, one federal appeals court has said yes. No court has applied that ruling to AI chats. In United States v. Warshak (December 14, 2010), the government had used a section 2703(b) subpoena and then a 2703(d) order to make an internet provider hand over about 27,000 of a man's emails, and he didn't learn of either until May 2006 (opinion).

The Sixth Circuit held that "if government agents compel an ISP to surrender the contents of a subscriber's emails, those agents have thereby conducted a Fourth Amendment search," and that "to the extent that the SCA purports to permit the government to obtain such emails warrantlessly, the SCA is unconstitutional" (Warshak). The emails were still used at trial, because the agents had relied on the statute in good faith. The ruling binds only federal courts in Kentucky, Michigan, Ohio, and Tennessee.

Warshak also left a door open. The court said "the mere ability of a third-party intermediary to access the contents of a communication cannot be sufficient to extinguish a reasonable expectation of privacy," but also that "a subscriber agreement might, in some cases, be sweeping enough to defeat" one (Warshak). OpenAI's privacy policy, as of November 2026, says it may use your content "to train the models that power ChatGPT" and lists "monitoring any Content submitted" among its purposes. No court has yet said whether terms like those are sweeping enough.

What does the Stored Communications Act require for ChatGPT records?

It depends on what the government wants, and the statute was written in 1986 for email and remote storage, long before chatbots. We matched each kind of request to the process 18 U.S.C. § 2703 and the case law set, read November 19, 2026.

What the government wantsWhat it needsWhere the rule comes from
Your name, address, session times, IP addresses, and how you payA subpoena, including one an agency writes itself, with no notice to you§ 2703(c)(2)
Other records about your account, such as login logsA court order on "specific and articulable facts" showing the records are "relevant and material"§ 2703(c)(1)(B) and (d)
Your chats, if the service counts as "electronic communication service" and they've been stored 180 days or lessA warrant§ 2703(a)
Your chats stored longer, or held by a "remote computing service"By the statute's text, a warrant, or a subpoena or 2703(d) order with notice to you that a court can delay. A warrant, in the Sixth Circuit§ 2703(b), Warshak
A freeze on your chats while police get the paperworkA request to the company: 90 days, plus 90 more on a renewed request§ 2703(f)
Location records over timeA warrant: getting them is a searchCarpenter, Chatrie
The user who typed a known prompt (a reverse search)A warrant has been used. Whether such warrants are constitutional hasn't been decidedMaine affidavit, CRS
The chat history on your own phone or laptopA warrant, even after an arrestRiley v. California

A chat can have strong protection while the account behind it has almost none, and the identity is often what police need first. In a February 2026 stalking case in Ohio, the FBI's affidavit says it received "2703(d) records from Google" showing "approximately 177,000 IP events" on the suspect's account before it got warrants for content (FBI affidavit).

Is ChatGPT an "electronic communication service" under the law?

No court has said. The statute defines "electronic communication service" as one that gives users "the ability to send or receive wire or electronic communications" (18 U.S.C. § 2510), and "remote computing service" as "the provision to the public of computer storage or processing services" (18 U.S.C. § 2711). A chat with a model isn't obviously a message to anyone, and processing a prompt is arguably computer processing.

The government hasn't picked one either. The first known affidavit for ChatGPT records calls OpenAI "an electronic communications service and/or remote computing service provider," and cites the warrant provisions for both (affidavit).

One clause turns on that answer. The remote computing rules in section 2703(b) cover content held "solely for the purpose of providing storage or computer processing services," and only "if the provider is not authorized to access the contents" for "any services other than storage or computer processing" (18 U.S.C. § 2703). A service whose terms let it use your chats to train its models is authorized to do more than store and process them. Whether that takes AI chats outside the clause is a question no court has answered.

Have police gotten warrants for ChatGPT prompts?

Yes, in at least three cases on the public record as of November 19, 2026.

  • Maine, 2025: OpenAI. A Homeland Security Investigations agent in Portland applied for a warrant to identify an unknown user of a dark-web child abuse site. The affidavit says the user told an undercover agent about "two unique, specific prompts" he'd typed into ChatGPT. One, which the warrant dates to about April 18, 2025, asked "what would happen if sherlock holmes met q from star trek?" Agents sent OpenAI a preservation request on May 20, 2025, and asked the court to delay notice to the user for 30 days (affidavit). The affidavit doesn't show when the warrant issued. The CRS sidebar says it was obtained in September 2025 and called for the user's names, account credentials, session histories, and IP addresses for April 2025.
  • Ohio, 2026: xAI. In a stalking case, the FBI says it "obtained and executed search warrants for Grok/x.ai" in January 2026, and that the return included "prompts and media used by the account" and a screenshot the suspect had uploaded minutes before one of his anonymous reports (affidavit filed February 20, 2026). The charges are allegations, not findings.
  • New York, 2026: OpenAI and Anthropic. In United States v. Richard Kim, a securities fraud case separate from Heppner, a magistrate judge issued warrants on May 21, 2026 for seven accounts Kim held at the two companies. When Kim moved to quash the OpenAI warrant, prosecutors replied that the Stored Communications Act "provides no authority for criminal defendants" to quash a warrant served on a provider, and that a user's remedy is to challenge the evidence after the search (government letter, June 17, 2026). The charges are allegations, not findings. As of that June 17, 2026 letter the case had not gone to trial, and the letter describes evidence the government "expects to introduce at trial."

The New York letter also says agents had already found "numerous prompts" Kim sent ChatGPT on his own devices: two laptops, five cellphones, and an iPad (government letter, June 17, 2026).

Can police use a keyword warrant to search everyone's AI prompts?

They've done it with Google searches, and two state supreme courts split on it. A reverse keyword warrant asks a company to name every user who typed certain words.

In People v. Seymour (October 16, 2023), Denver police asked Google for everyone who had searched for the address of a house where an arson killed five people, over the 15 days before the fire (opinion). The Colorado Supreme Court held that Seymour had a reasonable expectation of privacy in his Google search history under Colorado's constitution, which has long rejected the third-party doctrine. It let the evidence stand anyway, because the police acted in good faith, and made "no broad proclamation about the propriety of reverse-keyword warrants."

In Commonwealth v. Kurtz (December 16, 2025), Pennsylvania's supreme court reached the opposite result for "general, unprotected internet searches" (opinion). What matters, the lead opinion said, is that the user "is informed" that Google, "a third party," "will collect and store that information." Only two other justices joined it in full, and it left out searches where users "take additional steps to protect their privacy."

Kurtz came six months before Chatrie, whose majority rejected the argument that opting in to a Google service gives up privacy in it. No court has yet said how the two fit together, or applied either to a prompt sent to a chatbot.

Has any court ruled on privacy in AI chats?

Not on the Fourth Amendment question, as of November 19, 2026. The rulings so far are about privilege and civil discovery. In United States v. Heppner, a federal judge in New York held in February 2026 that a defendant's chats with Claude weren't privileged (memorandum). In the New York Times copyright case against OpenAI, a magistrate judge ordered OpenAI on November 7, 2025 to hand over a sample of 20 million de-identified ChatGPT logs, and the district judge affirmed in January 2026 (order). We cover both in can AI chats be subpoenaed. Whether a reverse search for a chatbot prompt is constitutional, whether AI terms of service defeat an expectation of privacy, and which part of the Stored Communications Act an AI service falls under are all undecided.

How Morse keeps your chat history in your browser

Each of these routes to a company runs through a copy the company keeps, and Chatrie and Warshak only decide how hard that copy is to get. A court can order a service to produce what it stores, so Morse is built not to store your chats. Your chat history is stored in your browser, encrypted, not on Morse's servers. Our servers don't keep messages or log message content, and your account holds 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 (Morse privacy).

Each chat also has a privacy mode that decides who handles the message on its way to the model:

  • E2EE, for your most sensitive chats. Before anything leaves your device, your browser checks the secure enclave's hardware attestation, then encrypts the message so only that enclave can read it, and Morse's servers relay text they can't read. 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. Models in E2EE can use only tools that run on your device, so there's no web search or image generation.
  • Private, the default, for everyday chats you'd rather no host kept. It runs open-weight models like DeepSeek, Kimi, and GLM. Morse has zero data retention agreements with its inference, search, and other providers, so they've agreed to 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 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 that would point back to you.

Your own device can still be searched. Police need a warrant to search a phone under Riley, agents in the Kim case found prompts on his own devices, and border agents need less, as we cover in can border agents see your ChatGPT. Sign out when you're done, since your history locks when you sign out, and keep your device locked. Clearing your browser or losing the device deletes your chats, and Morse can't recover them, so keep anything you need. Morse doesn't train AI on your chats, sell your data, or show ads.

Try a chat from the Morse home page without an account, or start free with no card. E2EE chats come with Pro, at $19 a month.

Common questions

Does the law let police get ChatGPT chats without a warrant?

The statute's text allows it for some stored content: section 2703(b) lets the government use a subpoena or a 2703(d) court order, with notice to you that a court can delay. The Sixth Circuit held in Warshak that email content needs a warrant, but that ruling binds only federal courts in Kentucky, Michigan, Ohio, and Tennessee, and no court has applied it to AI chats yet.

Does deleting a ChatGPT chat keep it away from police?

Not once police have asked the company to keep it. Under section 2703(f), a provider must preserve records for 90 days when the government asks, and another 90 on a renewed request. In the Maine case, agents sent OpenAI a preservation request on May 20, 2025, about a month after the prompt they were tracing.

Does the Fourth Amendment stop OpenAI from reading my chats?

No. The Fourth Amendment limits the government, not a company acting on its own, so a provider's own review of your chats is governed by its terms and privacy policy. It applies when the government compels the company to search or hand over your data.

Do state constitutions protect AI chats more than the Fourth Amendment?

Some do. Colorado's supreme court has long rejected the third-party doctrine under its own constitution and held in 2023 that a person has a reasonable expectation of privacy in their Google search history. Pennsylvania's supreme court went the other way on Google searches in December 2025.

Ask what you actually want to ask.

80+ models, characters, and images, with every chat encrypted and kept on your device. Free to start.

Start a private chat

Read next