AI

Can ChatGPT Chats Be Used in Court? What Rulings Show So Far

Courts have ordered AI chat logs preserved and produced, but the orders were narrower than the trending headlines suggest.

11 min read
Rows of tall black server racks with small blue indicator lights line both sides of a dim, gray, empty data center room under a row of ceiling lights.
Image by Schäferle from Pixabay

A federal court ordered OpenAI to keep ChatGPT logs and later to produce 20 million de-identified ones in a copyright case. Prosecutors have also cited chats in criminal cases, and courts disagree on whether chatbot conversations are privileged. Deleting chats or using Temporary Chat limits storage but cannot defeat a valid legal order.

Yes, in several ways. A federal court ordered OpenAI to preserve ChatGPT output logs and later to produce 20 million de-identified consumer conversations in a civil lawsuit. Prosecutors have cited chat prompts in criminal cases, and a federal judge in New York ruled that a defendant's conversations with a different chatbot, Anthropic's Claude, were not protected by attorney-client privilege.

What those rulings mean for an ordinary user is narrower than recent headlines imply. The two court orders in the best-known case were separate, and neither handed named individuals' chats to prosecutors or to ordinary litigants. Still, the cases show that chat logs are stored records, and stored records can be demanded. That makes the retention rules and the privilege question practical matters for anyone who types sensitive material into a chatbot.

Two orders, not one: what the New York Times case required

The trend that prompted this question, summarized by the aggregator AI Weekly, describes courts ordering AI chat copies kept and then handed over. In the New York Times lawsuit against OpenAI and Microsoft, which the case tracker AI Lawsuit Tracker dates to December 2023, those were two different orders with different reach.

The preservation order

On May 13, 2025, Magistrate Judge Ona Wang directed OpenAI to preserve and segregate all output log data that would otherwise have been deleted, as Thurrott reported in quoting the ruling. An explainer from the site LLMs for Lawyers says the obligation lasted until September 26, 2025, and that Enterprise customers were excluded. That is about 136 days during which deletion was effectively suspended for affected consumer chats.

A vendor blog quotes OpenAI as saying its obligations under that earlier order ended on September 26, 2025. The brief behind this article could not retrieve OpenAI's own page for the statement. A separate vendor blog claims that data from April to September 2025 remains in a restricted legal-hold environment and that some flagged users are still affected. That claim is unconfirmed.

The production order

On November 7, 2025, Judge Wang ordered OpenAI to produce 20 million de-identified consumer ChatGPT logs, according to PPC Land. The logs were a random sample of complete conversations from December 2022 to November 2024. The recipients were the news plaintiffs, who are OpenAI's opponents in the lawsuit. The sample excluded Enterprise, Edu, Business and API customers.

Thurrott's report says the judge found production relevant and proportional, noting that the sample was less than 0.05% of the total logs OpenAI retains in the ordinary course of business. eWeek reports that OpenAI holds "tens of billions" of consumer logs. Another explainer describes the 20 million as 0.5% of the preserved logs. That is a different denominator, and the two percentages should not be compared directly.

The size of the original request is disputed. The Decoder reports that the Times first sought 120 million chat logs and OpenAI offered 20 million. OpenAI's own account, relayed by a trade site, says the Times demanded 1.4 billion conversations in May 2025. The figures conflict, and OpenAI is a party to the case.

Privacy safeguards, as each side describes them

Reuters-based coverage in Malay Mail says OpenAI argued that 99.99 percent of the requested transcripts had nothing to do with the accusations. The Times countered that the court had ordered a sample of chats, anonymized by OpenAI itself, under a legal protective order. Judge Wang reportedly said users' privacy would be protected by "exhaustive de-identification" and other safeguards. Both parties' descriptions are partisan.

The order held up on review. According to the AI Lawsuit Tracker, District Judge Stein affirmed it on January 5, 2026. The LLMs for Lawyers explainer adds that he found privacy interests adequately balanced by de-identification and a protective order, though the order itself was not reviewed for this article.

The logs have since been used as evidence. PPC Land reports that OpenAI's own experts found 24 verbatim outputs in the 20 million produced logs, or 0.00012%, with the longest passages at 29 and 43 words. That finding is a party's expert claim from a single source, and the case is ongoing. Malwarebytes also reports that OpenAI had to keep data it had agreed to delete under the EU's GDPR and California's privacy laws, though that comes from a secondary source.

How chats reach investigators and prosecutors

Criminal cases involve different routes than civil discovery. According to the Cybersecurity Law Report's December 3, 2025 issue, a federal court in the District of Maine issued the first known federal search warrant asking OpenAI for user data. That investigation involved child exploitation.

The same report describes the Palisades fire case against Rinderknecht, in which the complaint used his ChatGPT prompts. He had consented in writing to a phone search and provided his PIN, so investigators did not need a warrant. The example shows that a chat history on a device can be reached without any request to the provider.

CNN, in a May 2, 2026 report, lists several further examples: the Palisades arson case, a Snapchat AI conversation in a 2024 Virginia murder trial, and a murder case (Abugharbieh) in which alleged ChatGPT entries were included in court documents. CNN also reports that Florida's attorney general opened a criminal investigation into OpenAI. That concerns the company's conduct, not chats as evidence.

A Colorado defense-lawyer blog lists the legal channels as a search warrant, a grand jury subpoena or a Stored Communications Act order, plus consent searches such as a suspect unlocking a phone. Another firm says content generally requires a warrant. Both are lawyer-marketing sources, and the statute and case law were not reviewed, so these descriptions need verification before being treated as settled.

Privilege: two rulings, no consensus

Attorney-client privilege shields confidential talks between lawyer and client from forced disclosure, and the work-product doctrine protects materials prepared for litigation. Whether either covers a person's chats with an AI tool is being decided case by case, and the early rulings point in different directions.

United States v. Heppner

In the Southern District of New York, Judge Rakoff held that written exchanges between a criminal defendant and Claude were protected by neither attorney-client privilege nor the work-product doctrine. The Harvard Law Review blog dates the government's motion to February 6, 2026, and the written opinion to February 17. The case did not involve ChatGPT, though the reasoning is being applied to consumer chatbots generally.

Law-firm analyses give two reasons. Claude is not an attorney, and the tool's policy reportedly disclaimed confidentiality. According to Jones Walker, the materials were 31 documents that the FBI seized in a search of the defendant's home. White & Case notes the ruling suggested protection might have been possible had counsel directed him to use the tool.

Harvard Law Review commentary argues that the opinion veers toward categorically excluding a client's use of generative AI from attorney-client privilege, and that this outcome was not inevitable. The ruling is a single trial-level decision.

Warner v. Gilbarco

On February 10, 2026, Magistrate Judge Patti in the Eastern District of Michigan reached a different result. Per Cleary Gottlieb, he held that a pro se civil litigant's ChatGPT queries and the AI's responses were protected work product. The court treated AI platforms as tools rather than persons. Fisher Phillips reports the court also found the materials irrelevant or not proportional.

The two rulings are not clean opposites, because the facts differ. One party was a represented criminal defendant acting without counsel's direction, and the other was a self-represented civil plaintiff. Mintz, in an April 29, 2026 analysis, says Warner and a third case, Morgan v. V2X, viewed AI as an ordinary drafting instrument, while Heppner treated consumer AI as a third party. The open disagreement, in its view, is whether attorney direction is required.

How long OpenAI says it keeps chats

OpenAI's help center page on chat and file retention is the main source for the company's rules. Deleting a chat removes it from the account view immediately, and OpenAI schedules permanent deletion within 30 days. The exception covers chats already de-identified and disassociated from the account. The page also says OpenAI may retain content longer for security or legal obligations, and that a deleted chat cannot be restored, even through support.

Regular chats stay until the user deletes them, and archiving does not change that. Temporary Chats do not appear in history, but OpenAI may keep a copy for up to 30 days for safety purposes. A temporary chat that is saved becomes a regular chat.

OpenAI's stated retention rules by content type, as of late 2026
Content Retention
Regular or archived chat Until the user deletes it
Deleted chat Gone from view at once; permanent deletion scheduled within 30 days (exceptions: de-identified, security, legal)
Temporary Chat Not in history; safety copy up to 30 days
API inputs and outputs Up to 30 days (some endpoint exceptions)
Enterprise and Edu Set by workspace admin; deleted conversations removed within 30 days unless legally required to keep them

The business tiers work differently. OpenAI's statement on the Times' data demands says Enterprise and Edu workspace admins control how long customer content is kept. Its enterprise privacy page says API data may be retained for up to 30 days to provide the services and identify abuse, with exceptions for certain endpoints. Both pages come from a party to the Times case.

The 2025 preservation order shows why the 30-day figure is not a guarantee. For roughly five months, a court order overrode the ordinary deletion schedule for affected chats. The policy and the order both existed, but whether any data from that period remains is unclear.

What users can do, and what it cannot do

The available controls reduce what is stored, but none defeats a valid legal order. Based on OpenAI's retention page and the surrounding reports, the main options are these:

  • Temporary Chat. Conversations stay out of history, though OpenAI may keep a safety copy for up to 30 days.
  • Deleting chats. The chat leaves the account view at once and is scheduled for permanent removal within 30 days, subject to legal and security exceptions.
  • Opting out of training. A secondary source, ai-toolbox.co, says personal-plan chats are used for training unless the user opts out and that Temporary Chats are not used. The exact opt-out steps were not checked against OpenAI's own data-controls page.
  • Business tiers. Enterprise and Edu admins set retention, and those customers were excluded from the 20 million-log sample.

Controls on OpenAI's servers also do nothing about copies on a person's own devices. The Palisades case turned on a consented phone search, not a request to the provider. Material that a user has saved, exported or left in a chat history on a phone can be reached by whoever lawfully holds the phone.

Nor do these steps alter the privilege picture. Under Heppner, a consumer's chats with a chatbot were not privileged, and no deletion setting changes how a court would classify them. This is researched context rather than personal or legal advice.

What the record leaves open

The New York Times order shows that a court can compel a provider to keep and produce user chats over the provider's privacy objections. It does not show that the Times or anyone else can read named individuals' chats. The sample was de-identified and produced under a protective order. Discovery in one case has not created a general rule, though it indicates that courts treat AI chats like other stored data.

Other chatbots may follow different retention rules. Anthropic, Google and Microsoft were outside the scope of the available research, and the GDPR conflict mentioned in the Malwarebytes post rests on that single secondary source.

What to watch

The clearest lesson so far is that a chat history is an ordinary record in the eyes of the courts. It can be preserved when a lawsuit is pending, sampled when a judge finds it relevant and proportional, and cited when prosecutors can lawfully reach it. The protections people may assume, such as confidentiality like a lawyer's or deletion that is final, are not guaranteed by the rulings reviewed.

The next developments to follow are an appellate ruling on whether consumer AI use can ever be privileged, a resolution of the attorney-direction question that Mintz identifies, and any official figures on government requests to AI providers. Until then, the law rests on a handful of district-court decisions that do not agree.

Frequently asked questions

Can ChatGPT conversations be subpoenaed?

Courts have compelled them in at least one civil case. A federal court ordered OpenAI to produce 20 million de-identified consumer logs in the New York Times lawsuit, and the order was affirmed on January 5, 2026. In criminal matters, a District of Maine court issued what is reported as the first known federal warrant to OpenAI for user data. The details of the legal routes come mostly from law-firm summaries.

Did OpenAI hand individual users' chats to the New York Times?

Not as reported. The produced logs were a random sample, de-identified and delivered under a protective order. They excluded Enterprise, Edu, Business and API customers. OpenAI objected to the order, and the Times said the sample was anonymized by OpenAI itself.

How long does ChatGPT keep deleted chats?

OpenAI's help center says a deleted chat disappears from the account view immediately and is scheduled for permanent deletion within 30 days. Exceptions apply to chats already de-identified and to content OpenAI must keep for security or legal reasons. A court order suspended deletion for affected chats from May 13 to September 26, 2025.

Are ChatGPT conversations protected by attorney-client privilege?

No general protection exists. In United States v. Heppner, a New York federal judge held that a defendant's Claude conversations were neither privileged nor work product. In Warner v. Gilbarco, a Michigan federal magistrate judge protected a self-represented litigant's ChatGPT use as work product. The two rulings involved different facts and are trial-level decisions.

Does Temporary Chat keep ChatGPT from storing a conversation?

Temporary Chats do not appear in history, but OpenAI says it may keep a copy for up to 30 days for safety purposes. The feature reduces storage but does not remove the chance of a legal demand during that period. A temporary chat that the user saves becomes a regular chat.

Sources (26)

  1. Chat and file retention in ChatGPT , OpenAI Help Center
  2. How we're responding to The New York Times' data demands in order to protect user privacy , OpenAI (2025-01-01)
  3. Enterprise privacy at OpenAI , OpenAI
  4. OpenAI Must Turn Over ChatGPT Logs in New York Times Case , Thurrott.com (2025-11-01)
  5. OpenAI Ordered to Hand Over 20M ChatGPT Logs , eWeek (2025-11-01)
  6. OpenAI ordered to turn over 20 million ChatGPT chats to the New York Times , The Decoder (2025-11-12)
  7. OpenAI resists court order to release more than 20 million ChatGPT logs in New York Times case , Malay Mail (2025-11-13)
  8. OpenAI quotes 20 briefs, US government included, against New York Times , PPC Land
  9. New York Times v. OpenAI & Microsoft - AI Lawsuit Tracker , ailawsuittracker.com
  10. NYT v OpenAI: The Deleted ChatGPT Chats Order, Explained , llms-for-lawyers.com
  11. United States v. Heppner , Harvard Law Review (2026-03-01)
  12. Heppner Shows Attorney-Client Privilege's Fragility in AI Era , Bloomberg Law
  13. Attorney-client privilege and work product in the age of generative AI , White & Case
  14. Your AI Conversations Are Not Privileged: What a New SDNY Ruling Means for Every Lawyer and Client , Jones Walker
  15. Your AI Conversations Are Not Privileged , frblaw.com
  16. Courts Grapple with Privilege Implications of AI , Cleary Gottlieb
  17. Can Your AI Chat History Be Used Against You in a Lawsuit? 5 Practical Takeaways for Employers as Courts Start to Split , Fisher Phillips
  18. Three Courts, No Consensus: The Evolving Privilege Landscape for GenAI-Generated Legal Materials , Mintz (2026-04-29)
  19. Dec. 3, 2025 issue , Cybersecurity Law Report (2025-12-03)
  20. How ChatGPT conversations became evidence in criminal investigations , CNN (2026-05-02)
  21. Anything You Ask ChatGPT May Be Used Against You in Court: The Dark Side Of AI , Colorado Criminal Lawyer Blog (2026-08-22)
  22. Your AI chats could be used in court , Malwarebytes (2026-09-01)
  23. AI News: Artificial Intelligence Stories, Ranked , AI Weekly (2026-11-05)
  24. Does ChatGPT Save Your Chats? Retention Explained (2026) , ai-toolbox.co (2026-01-01)
  25. OpenAI Data Retention Policy (2026): What They Keep, and For How Long , humla.team (2026-01-01)
  26. Does ChatGPT Store Your Data? Retention by Tier Explained , witness.ai (2026-01-01)

Share this article

Back to list