The New York Times and several other publishers filed a motion for serious sanctions against OpenAI on July 9, 2026, accusing the company of concealing key evidence for over two years in an ongoing copyright lawsuit. At the center of the allegations: OpenAI had tens of millions of anonymized ChatGPT output logs — and never disclosed them to the court, while simultaneously claiming that searching such data was technically infeasible.
Key takeaways
- NYT is seeking sanctions over OpenAI hiding samples of 10M and 78M ChatGPT logs for more than two years
- OpenAI had previously told the court that searching such logs was technically impossible and burdensome
- OpenAI privacy engineer Vincent Monaco revealed in an April 2026 deposition that the company had previously searched those logs itself
- The 20M-log sample given to publishers was deemed "unusable" by the court after 19 billion AI-generated redactions
- Sanctions could block OpenAI from making key fair use arguments
Two years of hidden evidence
The NYT v. OpenAI and Microsoft case has been running since late 2023. Publishers allege that ChatGPT was trained on millions of protected articles and can reproduce their content verbatim on demand — constituting copyright infringement and undermining their subscription businesses.
Central to the dispute are ChatGPT output logs — records of the model's responses that could reveal whether and how often the chatbot reproduced NYT articles word for word. Publishers demanded access to such data from the start. OpenAI spent two years arguing that searching the logs was technically difficult, expensive, and risked user privacy.
The turning point came during a second deposition of privacy engineer Vincent Monaco in April 2026. He testified that OpenAI had already possessed two anonymized samples — 10 million and 78 million records — before litigation began. The company had also already searched those samples to build a filter for blocking the reproduction of copyrighted content.
OpenAI was willing and able to search its output logs — when it benefited OpenAI
NYT, sanctions motion, July 9, 2026
An "unusable" sample and billions of deleted records
Instead of disclosing the existing samples, OpenAI provided publishers with a different dataset — 20 million logs in a "sandbox," with access limited to a heavily redacted version. The company applied AI to make 19 billion redactions, including removing publisher domain names and identifiers. The court ultimately found that sample "unusable".
Beyond hiding existing samples, NYT also accuses OpenAI of deleting or compressing billions of logs that should have been preserved under a court preservation order. According to Monaco's testimony, OpenAI "decided" that complying with the order "would be hard" and took no steps to do so.
What sanctions NYT is requesting
Publishers are asking the court for three categories of sanctions. First, prohibit OpenAI from using the 20 million sample in proceedings. Second, declare as established fact that the withheld logs contained "substantial" verbatim reproduction of protected content. Third, instruct the jury that OpenAI deleted billions of logs — feeding directly into publishers' narrative that OpenAI has obstructed evidence.
OpenAI disputes the allegations, arguing the sanctions motion is an attempt to force access to private user data and that the NYT case is weakening. NYT lead counsel Ian Crosby pushed back: "For over two years, OpenAI lied to The Times, the Daily News Plaintiffs, the public, and the court. It claimed searching ChatGPT outputs was infeasible — while concealing that it had already done such searches."
Why this matters
The outcome of this sanctions motion could determine the outcome of the entire case. OpenAI's defense rests heavily on the fair use?fair use: A US copyright doctrine allowing limited use of protected content without the owner's permission, including for transformative purposes. argument — that training on publicly available material is permissible because it does not harm the market for the original content. Establishing the actual frequency of verbatim reproduction is critical to proving or disproving that claim.
If the court grants sanctions and limits OpenAI's ability to contest the scope of infringement, the company loses one of its most important defensive tools. The broader context: the NYT case is being watched across the entire AI industry. A win for publishers could open the door to similar suits against other companies training models on protected data.
For OpenAI, the stakes extend beyond this case. The company is approaching a prospective IPO, and a court finding of deliberate evidence concealment would be a serious blow to its reputation and valuation.
What's next?
- The court must rule on the sanctions motion — NYT expects OpenAI's response within weeks
- If sanctions are granted, OpenAI will lose use of the 20M log sample and likely the ability to contest the scope of infringement
- The main trial has no set date, but the sanctions ruling could significantly shift the negotiating dynamics for both sides
Sources
- Ars Technica — OpenAI may have made a fatal misstep in copyright fight with news orgs
- Ars Technica (archive) — OpenAI says court forcing it to save all ChatGPT logs is a privacy nightmare
