Yes, and they already have been. In New York Times v. OpenAI, a court ordered OpenAI to produce 20 million de-identified ChatGPT conversations. Judge Sidney Stein affirmed that order on 5 January 2026. A separate preservation order forced OpenAI to retain deleted chats; its obligations under that order ended 26 September 2025.
The short answer, from OpenAI's own privacy policy
OpenAI's privacy policy states the position directly. Under Retention, it says: "If we are legally required to retain your data (for instance, we receive a lawful subpoena) then we may retain it for the duration of the relevant legal or regulatory obligation" (OpenAI Privacy Policy, section 4).
The disclosure section is broader still. OpenAI may share personal data, "including information about your interaction with our Services," with government authorities or other third parties "if required to do so to comply with a legal obligation," among other grounds (OpenAI Privacy Policy, section 3).
So the answer is not theoretical. Your conversations are records held by a third party, and records held by third parties are reachable by legal process.
This page is not legal advice. It is a factual account of a public court docket and the vendor's published terms.
Correcting the record: the preservation order ended in 2025, not 2026
A great deal of secondary coverage places the lifting of the ChatGPT preservation order in October 2026. That is wrong, and the primary source is OpenAI's own case page.
OpenAI's update, dated 22 October 2025, states: "After months of litigation, we are no longer under a legal order to retain consumer ChatGPT and API content indefinitely. Our obligations under the earlier order ended on September 26, 2025" (OpenAI, response to NYT data demands).
The date that gets misremembered is the order itself, filed by Magistrate Judge Ona T. Wang on 9 October 2025, which freed OpenAI from the obligation to preserve and segregate output log data going forward. October 2025. Not 2026.
What the preservation order covered, and who it missed
While it ran, the preservation order was broad. It required OpenAI to keep chats that its normal 30-day deletion cycle would have removed, including chats users had deliberately deleted.
OpenAI listed the impact by plan. Affected: ChatGPT Free, Plus, Pro and Team subscribers, and API users without a Zero Data Retention agreement. Not affected: ChatGPT Enterprise, ChatGPT Edu, and API customers using Zero Data Retention endpoints under OpenAI's ZDR amendment (OpenAI, response to NYT data demands). OpenAI notes the court clarified on 27 May that ChatGPT Enterprise was excluded from preservation.
Even after the order ended, a residue remains. OpenAI says the Times "continues to demand that OpenAI keep a specific set of user data from April-September 2025," which OpenAI stores under legal hold, "accessible only to a small, audited OpenAI legal and security team," and which "will not be turned over to the New York Times, the Court, or anyone else at this time" (OpenAI, response to NYT data demands).
The bigger event was production, not preservation
Preservation means keep. Production means hand over. The second one happened.
In July 2025 the plaintiffs moved to compel OpenAI to produce a sample of 120 million chat logs. OpenAI opposed it and offered a sample of 20 million instead, scrubbed of personally identifiable information. The plaintiffs agreed to the 20 million while reserving the right to seek more (ABA Journal, 8 January 2026).
OpenAI later tried to narrow it, proposing keyword searches to produce only conversations implicating the plaintiffs' works. Magistrate Judge Ona Wang rejected that. On 5 January 2026, US District Judge Sidney Stein of the Southern District of New York affirmed her orders, writing that "Judge Wang's rulings were neither clearly erroneous nor contrary to law" and that "she adequately balanced ChatGPT users' privacy interests against the relevance of the documents in light of the privacy protections already in place" (ABA Journal, 8 January 2026).
The three safeguards the court relied on were the cap at 20 million logs, OpenAI's de-identification of the sample, and the existing protective order in the case.
Whose conversations, and who gets to read them
OpenAI has published the sampling parameters. "The 20 million user conversations were randomly sampled from Dec. 2022 to Nov. 2024," and conversations outside that window are not affected (OpenAI, fighting the NYT's invasion of user privacy).
Business tiers are out. OpenAI states the production "does not impact ChatGPT Enterprise, ChatGPT Edu, ChatGPT Business (formerly 'Team') customers, or API customers" (OpenAI).
Access is limited but real. "The Times' outside counsel attorneys of record in the case and their hired technical consultants would be able to access the conversations," OpenAI says, adding that it would push for viewing inside a secured environment under strict legal protocols (OpenAI). OpenAI also notes that its original exposure was far larger: the Times initially demanded 1.4 billion conversations.
Where the internet is commonly wrong about the 20 million
Coverage almost universally reports that the Times demanded 20 million conversations. OpenAI's own blog says exactly that: "The New York Times is demanding that we turn over 20 million of your private ChatGPT conversations" (OpenAI).
The docket tells a more awkward story. Per the ABA Journal's account of Judge Stein's order, the plaintiffs moved for 120 million logs, and it was OpenAI that offered the 20 million sample as a counter (ABA Journal, 8 January 2026). OpenAI then sought to narrow its own offer, and the court held it to the number.
Both sources are worth reading together. The number was not imposed on OpenAI from nowhere. It was OpenAI's proposal, later regretted.
Here is how the two orders landed across the plans:
| Plan | Covered by the 2025 preservation order | Covered by the 20-million production sample | Standard deletion |
|---|---|---|---|
| ChatGPT Free / Plus / Pro | Yes | Yes, if sampled from Dec 2022 – Nov 2024 | Within 30 days |
| ChatGPT Business (formerly Team) | Yes | No | Within 30 days |
| ChatGPT Enterprise | No | No | Admin-set |
| ChatGPT Edu | No | No | Admin-set |
| API (standard) | Yes | No | After 30 days |
| API with Zero Data Retention | No | No | Never logged |
Preservation and deletion columns from OpenAI's response to the NYT data demands; production column from OpenAI's user privacy post.
Where the case stands as of 3 September 2026
The litigation is active in the Southern District of New York, now part of a consolidated multi-district proceeding involving multiple news organizations. Judge Stein denied the motions to dismiss in large part on 26 March 2025, allowing the central copyright claims to proceed.
In July 2026 the Times sought sanctions against OpenAI, alleging the produced output logs showed "substantial and systematic grounding on and regurgitation of News Plaintiffs' copyrighted material" (Bloomberg Law). That allegation only exists because the logs were produced.
Most recently, Judge Stein ordered the plaintiffs to show cause in writing by 11 September 2026. The question is whether the action should be stayed pending summary judgment motions in other MDL cases. The defendants' response is due 18 September 2026. No trial date has been set. This status is current as of 3 September 2026 and will move.
Seven things to do about it
- Assume every conversation is a record. Write in an AI chat the way you would write in a company email that might be read out in court.
- Redact before you paste. Names, account numbers, case numbers, exact figures. A conversation with no identifiers is a conversation with limited evidentiary value against you.
- Move sensitive work to a business tier. Enterprise, Edu and ZDR API were outside both the preservation and the production orders.
- Do not treat deletion as destruction. OpenAI deletes within 30 days, "unless we need to retain it for longer" for legal reasons (OpenAI Privacy Policy). A legal hold overrides your delete button.
- Use Temporary Chats for one-off sensitive questions. They stay out of history and are deleted within 30 days (OpenAI Data Controls FAQ).
- Fold AI tools into your litigation hold process. If your firm issues holds, the list of systems should name every AI tool your staff use.
- Ask your vendor for the retention period in days, in writing. Marketing pages change. A contractual number does not.
The line worth remembering
A cybersecurity lawyer commenting on the ruling put it more cleanly than any policy document. "This case is also a telling reminder that — regardless of your privacy settings — your interactions with AI chatbots and other systems may, one day, be produced in court," Ilia Kolochenko of Platt Law told Law.com (quoted by the ABA Journal, 8 January 2026).
Privacy settings govern what a vendor does voluntarily. They do not govern what a court orders.
sources:
- OpenAI — How we're responding to The New York Times' data demands (update dated 22 October 2025): https://openai.com/index/response-to-nyt-data-demands/
- OpenAI — Fighting the New York Times' invasion of user privacy: https://openai.com/index/fighting-nyt-user-privacy-invasion/
- OpenAI — Privacy Policy, sections 3 and 4: https://openai.com/policies/privacy-policy/
- OpenAI Help Center — Data Controls FAQ (Temporary Chats): https://help.openai.com/en/articles/7730893-data-controls-faq
- OpenAI Help Center — Chat and file retention policies in ChatGPT: https://help.openai.com/en/articles/8983778-chat-and-file-retention-policies-in-chatgpt
- ABA Journal — ChatGPT creator must turn over 20M chat logs in copyright litigation, federal judge says (8 January 2026): https://www.abajournal.com/news/article/chatgpt-creator-must-turn-over-20m-chat-logs-in-copyright-litigation-federal-judge-says
- Bloomberg Law — New York Times Seeks Sanctions Against OpenAI in Copyright Case: https://news.bloomberglaw.com/ip-law/new-york-times-seeks-sanctions-against-openai-in-copyright-case
- Justia Dockets — The New York Times Company v. Microsoft Corporation et al, preservation order: https://docs.justia.com/cases/federal/district-courts/new-york/nysdce/1:2023cv11195/612697/551
- OpenAI Developer Docs — Data controls in the OpenAI platform (30-day abuse log retention, ZDR): https://developers.openai.com/api/docs/guides/your-data
related:
- Is it safe to put my client data into ChatGPT?
- Does deleting a ChatGPT conversation actually delete it?
- Can lawyers use ChatGPT and keep client confidentiality?
- What is Zero Data Retention and how do I get it?
