Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Fix the driver behind crashes, sound loss and screen glitches3Clear out junk files and repair common Windows errorsShort answer: The New York Times and other news plaintiffs allege that OpenAI deleted or failed to preserve evidence relevant to their copyright case. But the public record involves several different events—not one confirmed destruction of “all the evidence.” They include a 2024 deletion in a search environment, routine deletion of ChatGPT output logs, the 2022 deletion of training datasets called Books1 and Books2, and a 2026 request for sanctions. Courts have ordered OpenAI to preserve and produce data, but the record available through August 18, 2026, does not establish a final finding that OpenAI intentionally destroyed evidence.
What the evidence dispute is about
The underlying case began when The New York Times sued OpenAI and Microsoft on December 27, 2023. The Times alleged that its journalism was used to train AI systems without permission and that chatbot responses could reproduce or substitute for Times content. The lawsuit concerns both copyright liability and what evidence can show how OpenAI’s systems were trained and how they generated outputs. The Associated Press reported on the original lawsuit.
| # | Preview | Product | Price | |
|---|---|---|---|---|
| 1 |
|
Copyright Law | $149.28 | Buy on Amazon |
| 2 |
|
Copyright Law: Cases and Materials (v8.0) | $21.70 | Buy on Amazon |
| 3 |
|
Copyright Law of the United States: and Related Laws Contained in Title 17 of the United States Code | $10.32 | Buy on Amazon |
| 4 |
|
Copyright Law in a Nutshell | $65.00 | Buy on Amazon |
| 5 |
|
Copyright Handbook, The: What Every Writer Needs to Know | $37.99 | Buy on Amazon |
As an Amazon Associate I earn from qualifying purchases.
The preservation fight is separate from the copyright merits. Evidence that a model produced text resembling a news article may be relevant, but it does not by itself decide whether training or a particular output infringed copyright, whether fair use applies, or whether Microsoft has independent liability.
Free tools Windows power users keep installed
One-click scans. No signup required.
There are several different deletion disputes
| Data or event | When | What is disputed |
|---|---|---|
| Search-environment files and directory information | 2024 | Whether an engineering change deleted folder structure and file names in a temporary cache and impaired the plaintiffs’ searches. |
| Consumer ChatGPT output logs | Routine practice and later litigation preservation | Whether logs that would normally be deleted should have been retained because they could show reproduction of copyrighted material. |
| Books1 and Books2 datasets | Approximately 2022 | Whether deleted training datasets could have been relevant evidence. The deletion predates the lawsuits. |
| ChatExplorer logs | Disputed in the litigation | OpenAI says the Times itself deleted evidence relating to its internal use of OpenAI systems. |
| De-identified ChatGPT conversations | 2025 onward | How much data OpenAI must produce while protecting users’ privacy. |
The 2024 accidental-deletion incident
The News plaintiffs were given access to virtual machines so they could search OpenAI’s training data for their copyrighted works. In 2024, OpenAI engineers allegedly made a change that removed folder structure and file names from one drive used in that process.
#1 Best Overall
OpenAI described the drive as a temporary cache and said there was no reason to believe the underlying files themselves were lost. The Times and the other plaintiffs nevertheless argued that the deletion could impair their ability to identify and investigate relevant material. That distinction matters: metadata and directory information can affect what investigators can find even if copies of underlying files remain elsewhere.
TechCrunch reported on the 2024 incident. It should not be summarized as a final judicial finding that OpenAI intentionally destroyed evidence.
Why ordinary ChatGPT deletion became a legal issue
OpenAI has said that, under its normal policy, users can delete chats and deleted conversations are generally scheduled for permanent deletion within 30 days, subject to exceptions involving legal obligations, security and other circumstances. The Times argued that output logs could show whether ChatGPT reproduced or referenced copyrighted news content.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchPC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11That creates a conflict between two different systems:
- Product retention: a user deletes a chat and the service follows its ordinary deletion schedule.
- Litigation preservation: once relevant evidence must be preserved, a company may have to retain data that its normal policy would otherwise remove.
A user’s deletion request is therefore not automatically the final word if a court preservation obligation applies. Conversely, a preservation order does not automatically mean that OpenAI previously acted in bad faith.
What the May 13, 2025 order required
On May 13, 2025, Magistrate Judge Ona T. Wang ordered OpenAI to preserve and segregate output-log data that would otherwise have been deleted, including data subject to user deletion requests, until further order. The order also addressed privacy concerns, including whether data could be separated or anonymized in a way that reduced the impact on users.
The order was prospective. Its central purpose was to prevent additional loss while the parties’ discovery dispute continued. It was not, by itself, a ruling that OpenAI had intentionally destroyed evidence or that the Times had already proved its copyright claims. Read the May 13, 2025 order.
Recommended Free Tools
What plaintiffs wanted the logs to show
The News plaintiffs sought output logs and related information that could help them determine:
Rank #3
- whether OpenAI models generated memorized or near-verbatim passages from Times and other news articles;
- how frequently the models produced plaintiffs’ copyrighted material;
- whether OpenAI searched, or could search, its models and datasets for the plaintiffs’ works;
- whether deleted or withheld records affected the plaintiffs’ ability to prove their claims; and
- whether OpenAI complied with preservation obligations.
The court later addressed production of large volumes of de-identified logs. One order involved 20 million ChatGPT logs, while the record also refers to larger reservoirs of 78 million and 10 million logs. The relevant court order explains the production and privacy framework.
The separate Books1 and Books2 dispute
OpenAI deleted the Books1 and Books2 datasets in approximately 2022, according to the court record. These were training-data datasets, not user conversation logs. Their deletion occurred before the Times lawsuit and before the other News plaintiffs’ litigation.
That timing is legally important. A dataset deleted before a lawsuit or preservation obligation existed cannot simply be treated as a post-lawsuit violation of a litigation hold. The court’s November 24, 2025 opinion addressed the deletion and related privilege questions, including issues involving relevance and possible alternative evidence. Read the court’s discussion of Books1 and Books2.
The existence of the deletion may still matter to discovery. The parties can dispute what the datasets contained, whether other evidence can establish their contents or use, and whether the loss caused meaningful prejudice. But it is inaccurate to describe the 2022 deletion as though it necessarily violated a preservation order issued later.
Rank #4
Why user privacy is part of the dispute
The requested evidence includes conversations created by ordinary ChatGPT users. Preserving and producing millions of chats can affect people who are not parties to the copyright case and who may have expected deleted conversations not to remain available.
OpenAI has argued that indefinite retention and disclosure could undermine user deletion choices and privacy commitments. The court’s approach has included de-identification and protocols intended to remove personally identifying and other private information before production. De-identification can reduce exposure, but it does not guarantee that every person or sensitive detail is impossible to identify.
This is not simply a dispute over computer storage. The court must balance the plaintiffs’ need for potentially relevant evidence against the privacy interests of users whose conversations may be swept into discovery. OpenAI’s explanation of its retention and privacy objections is available here.
What happened in 2025 and 2026
- December 2, 2025: OpenAI was directed to produce 20 million de-identified ChatGPT logs. See the order.
- December 29, 2025: The News plaintiffs filed a supplemental memorandum alleging broader deletion of logs and seeking continued preservation measures. Read the supplemental memorandum.
- July 9, 2026: The Associated Press reported that the Times, Daily News and other plaintiffs sought sanctions against OpenAI over alleged discovery misconduct.
A request for sanctions is not the same as sanctions being imposed. As of the public record summarized here, the plaintiffs’ allegations remain contested and the reporting does not establish a final sanctions judgment.
Best Value
What OpenAI says
OpenAI disputes the Times’ characterization of the events. Its public account says the Times falsely accused it of destroying data, emphasizes the privacy consequences of retaining users’ conversations, and argues that the Times itself deleted evidence connected with its use of OpenAI models.
OpenAI’s position also distinguishes between routine deletion and intentional destruction. It has said that the 2024 engineering incident affected folder structure and file names in a temporary cache, not necessarily the underlying files. OpenAI’s public response to the Times’ claims is available here.
Did the court find that OpenAI spoliated evidence?
Not on the public record covered here as a final finding of intentional destruction. The court has addressed preservation concerns, ordered data to be retained, and directed production of substantial volumes of de-identified logs. Those orders show that the evidence dispute is serious and consequential.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →They do not automatically establish that OpenAI acted in bad faith. In a spoliation dispute, the court generally must examine questions such as:
- when a preservation duty arose;
- what data was within the party’s possession, custody or control;
- whether the loss was routine, accidental, negligent or intentional;
- whether the evidence could be recovered from another source; and
- whether the loss caused relevance, prejudice or other harm.
The public record therefore supports precise wording: the Times alleges deletion or withholding of relevant evidence; OpenAI disputes that account; and the court has imposed preservation and production requirements. It does not support presenting intentional destruction as an already adjudicated fact.
What remains unresolved
- Whether OpenAI intentionally destroyed or withheld any relevant evidence.
- Whether the 2024 deletion materially impaired the plaintiffs’ searches.
- Whether any lost data can be reconstructed or replaced with other evidence.
- Whether the plaintiffs suffered legally significant prejudice.
- Whether the court will impose sanctions in response to the 2026 request.
- Whether OpenAI’s training practices or particular outputs infringed copyright.
- How much user data can be disclosed while preserving meaningful privacy protections.
The practical takeaway
The most accurate summary is not “OpenAI deleted the evidence.” It is that the Times and allied plaintiffs are pursuing a serious but unresolved discovery and preservation claim involving different kinds of data and different dates.
The 2024 cache incident, ordinary ChatGPT deletion, the 2022 Books1 and Books2 deletion, and the 2026 sanctions request should be analyzed separately. The court’s preservation order means that ordinary product deletion rules may not apply once a legal preservation obligation attaches, but the order itself is not a finding that OpenAI intentionally destroyed evidence.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
