The ChatGPT court order on data retention: what it kept, and what it still holds

The Morse teamPublished October 19, 20268 min read

The ChatGPT court order on data retention no longer applies to new chats. As of October 2026, OpenAI is back to deleting deleted and Temporary chats within 30 days, because its duty to keep ChatGPT logs it would otherwise delete ended on September 26, 2025. Two pieces of that fight remain. OpenAI still holds a locked set of chats from April to September 2025, and a separate order made it give the New York Times and other plaintiffs a sample of 20 million de-identified ChatGPT conversations, which the Times is now asking the court to sanction OpenAI over.

On October 19, 2026, we read the orders and OpenAI's August 2026 court brief on CourtListener and Justia, the docket of In re OpenAI, Inc. Copyright Infringement Litigation (S.D.N.Y., 25-md-3143), and OpenAI's two posts on the Times' demands. Where a fact comes only from news reporting, we say so. This isn't legal advice.

Does OpenAI still keep deleted ChatGPT chats?

Not under a court order, for chats since September 26, 2025; we lay out the main rulings and the 30-day rule in can AI chats be subpoenaed. What's left is narrow. OpenAI keeps preserving the logs of accounts tied to 89 web domains the news plaintiffs listed in the stipulation, and OpenAI says the historical April to September 2025 data "will not be turned over" to the Times, the court, "or anyone else at this time." As of October 19, 2026, that October 2025 update is still the latest on OpenAI's page.

What did the preservation order actually keep?

Less than "every chat, forever." OpenAI's opposition brief of August 14, 2026 describes what it did after Magistrate Judge Ona T. Wang's May 13, 2025 order. It set aside, in "a segregated legal-hold database, all consumer ChatGPT logs generated between April 14, 2025 (i.e., 30 days prior to the Order) and September 26, 2025," including deleted chats and Temporary Chats. A footnote adds that the held data "includes both consumer and API chats."

Older chats were treated differently. OpenAI told the court it could not change the retention rules of its compressed warehouse of saved chats, so "users could thus still delete chats that were more than 30 days old as of the date of the Preservation Order." According to the brief, Judge Wang clarified at a May 27, 2025 hearing that the order did not require "wholesale" preservation or an end to all "deletion processes."

The order itself had grown out of a January 2025 conference, where Judge Wang declined the Times' request for "wholesale preservation of the output log data" but asked whether chats users wanted deleted could be set aside. OpenAI objected to the May order, and Judge Sidney H. Stein denied the objection on June 26, 2025, "for the reasons set forth on the record," after oral argument (ECF 271).

Who was covered by the order, and who wasn't?

ChatGPT's consumer plans and most API use were covered, while Enterprise, Edu, and Zero Data Retention customers were not. OpenAI's June 2025 post answers "Is my data impacted?" plan by plan, and its later post on the 20 million logs does the same for the sample.

OpenAI's explanation for the Zero Data Retention row is the shortest: "we never retain the prompts you send or the answers we return. Because it is not stored, this court order doesn't affect that data."

How did 1.4 billion chats become a 20 million-log sample?

Through a year of bargaining over what OpenAI already stores. OpenAI's security chief, Dane Stuckey, wrote that the Times first wanted 1.4 billion conversations. Judge Stein's January 5, 2026 order records the steps after that. In July 2025 the news plaintiffs moved to compel 120 million consumer logs, and OpenAI offered 20 million, saying that was "surely more than enough." The plaintiffs accepted the 20 million on August 11, 2025, while still asking for more. On October 14, OpenAI said it would run search terms over the sample instead of producing all of it, and the next day the plaintiffs moved to compel the whole sample. Judge Wang ordered it produced on November 7, and to the class plaintiffs on December 5. Judge Stein affirmed; we cover his privacy reasoning in can AI chats be subpoenaed.

The sample came from kept chats, not deleted ones. It was drawn at random from consumer conversations between December 2022 and November 2024, before the preservation order existed, out of what the court calls the "tens of billions" of logs OpenAI retains "in the ordinary course of its business." OpenAI's brief puts its compressed store at "50+ billion" output logs. The end of the preservation order stopped OpenAI keeping chats people deleted. It did nothing for the chats people kept, and those were the ones sampled.

In the same post, Stuckey wrote that OpenAI's long-term roadmap includes "client-side encryption for your messages with ChatGPT," so conversations would be "inaccessible to anyone else, even OpenAI."

Why is the Times seeking sanctions over the sample?

The Times says the sample can't be trusted, and OpenAI says users' own deletions explain the gap. In July 2026 the Times and Daily News asked the court to bar OpenAI from using the 20 million logs and to pay their fees, TechCrunch reported; they allege OpenAI deleted logs it should have kept and over-redacted the rest. These are allegations, and OpenAI denies them.

OpenAI's August 14 brief gives its side. It could retrieve "95%" of the 20 million logs the plaintiffs picked. The other 5% "were the subject of manual deletion requests by ChatGPT users" in the months before the plaintiffs chose their sample in August 2025, so OpenAI swapped in replacements, which the plaintiffs elected to have picked "at random." OpenAI calls the sanctions the Times wants "an unjustified windfall."

The news plaintiffs filed their reply on September 10, 2026. As of October 19, the docket shows the case in summary judgment briefing and no ruling on sanctions.

The order, date by date

The milestones other pages cover, the May 13, 2025 order, the September 26 end date, the November 7 sample order, and Judge Stein's January 5, 2026 ruling, are in our subpoena timeline. These are the rest, from the filings and OpenAI's posts.

For what OpenAI does with the chats it keeps outside of court, see does ChatGPT train on your data.

How Morse keeps your history out of a court order

Both orders reached ChatGPT conversations because OpenAI stored them. A court can order a service to produce what it stores, so Morse is built not to store your chats. Your history lives in your browser, encrypted, and our servers don't keep messages or log message content.

• No server copy to preserve or sample. A preservation order tells a company to stop deleting, and a sample order tells it to hand over part of what it kept. Morse's account holds your sign-in details, plan, balance, and a usage ledger with no content, so neither has a chat log to work on. Our privacy policy puts it this way: "Because we don't store your chats, we usually have very little to give."
• Deleting means deleted. Clear your browser or delete your account and your chats are gone, with no server copy waiting out a 30-day window. Morse can't recover them either, so keep anything you need.

Then pick a privacy mode per chat, depending on who should handle the message on its way to the model:

• Private, for chats no host should keep. It's the default. Open-weight models like DeepSeek, Kimi, and GLM, plus Grok, run under zero data retention: Morse has zero data retention agreements with its inference, search, and other providers, so they keep nothing. Where a provider confirms that on each reply, Morse checks it and refuses a reply that doesn't. It's the same reason OpenAI gave for leaving its Zero Data Retention customers out of the order.
• E2EE, for anything no one should read. Your browser checks the secure enclave's proof that it's genuine, then encrypts the message so only that enclave can read it. Our servers relay ciphertext they can't read. The enclaves are run by third-party providers, not by Morse, and the guarantee rests on the enclave hardware and attestation working as designed. Read how end-to-end encrypted AI chat works.
• Anonymous, to use GPT, Claude, or Gemini without the provider knowing it's you. Our proxy strips your account, email, and IP address and sends the message under Morse's own account. The provider may keep it for a limited time to check for abuse, under its own policy, so for anything you'd keep out of every company's logs, use Private or E2EE.

In every mode, Morse doesn't train AI on your content, sell your data, or show ads. Your stored history takes two keys to unlock: one that never leaves your device, and one our server gives you when you sign in. The copy on your device is the one to protect, so lock the device and sign out when you step away.

• If you're worried about an old ChatGPT chat: deleting it now removes it from your account, but not from a legal hold. Don't count on deletion during a dispute; ask a lawyer first.
• If you want new chats out of any company's logs: use a service that doesn't keep them. On Morse, start the chat in Private or E2EE.

Try a chat from the Morse home page without an account, or start free. End-to-end encrypted chats come with Pro, at $19 a month.