Analysis
The New York Times and the Daily News asked a federal judge on July 9 to sanction OpenAI, alleging the company misrepresented its own technical capabilities during discovery in the two-year-old copyright lawsuit over ChatGPT's training data. The filing claims OpenAI told the court it couldn't search its training corpus or chat logs for copyrighted journalism, when the company had already built and used tools that did exactly that.
The most damaging detail comes from an April 2026 court-ordered deposition of OpenAI data-privacy engineer Vinnie Monaco, who reportedly disclosed that OpenAI had already conducted internal searches of its training data for copyrighted news content, and had separately amassed a database of roughly 78 million de-identified ChatGPT conversations under an internal initiative called Project Giraffe. That effort reportedly included a tool called a "Bloom" filter, built specifically to detect and record when ChatGPT outputs were regurgitating source material -- built, plaintiffs say, shortly after the lawsuit was filed, undercutting any claim that OpenAI lacked the technical means to search for infringement at scale.
The evidence-handling allegations go further than a capability dispute. Plaintiffs say OpenAI deleted billions of ChatGPT outputs after the court issued a preservation order specifically requiring the company to retain that data, and that OpenAI substituted millions of logs within the sample it was ultimately ordered to produce. The scope of that production was itself a fight: the Times and Daily News originally sought 120 million chat logs, and after negotiation received 20 million -- a sample the plaintiffs say arrived so heavily redacted that the court itself has called it effectively unusable as evidence.
“Any one of those sanctions would materially shift leverage in a case that has run since December 2023 and names Microsoft as a co-defendant alongside OpenAI.”
The requested remedy is unusually aggressive for a discovery dispute. Plaintiffs are asking the judge to exclude the 20 million-log sample entirely, to instruct the jury that regurgitation of copyrighted content can be treated as an established fact rather than something OpenAI gets to contest at trial, and to award legal fees tied to the alleged misconduct. Any one of those sanctions would materially shift leverage in a case that has run since December 2023 and names Microsoft as a co-defendant alongside OpenAI.
OpenAI has pushed back hard. Spokesperson Drew Pusateri said in a statement that as the Times' underlying case weakens, the plaintiffs are using the discovery fight to try to access the private conversations of ChatGPT users who have nothing to do with the litigation, and the company continues to argue its use of copyrighted material falls under fair use. That framing -- privacy concerns versus evidence-preservation obligations -- is likely to become the central fight in the sanctions motion itself.
The case sits alongside a growing docket of AI copyright litigation, including suits from authors, visual artists and other publishers against OpenAI, Anthropic, Google and Meta, most of which turn on the same underlying question of whether training on copyrighted material without a license constitutes fair use. A sanctions finding against OpenAI wouldn't resolve that question directly, but it would set a discovery-conduct precedent other plaintiffs' lawyers are likely to cite immediately in parallel cases.
For AI founders building products on top of any frontier lab's API, the case is a pointed reminder that discovery obligations around user data retention are becoming a real, litigable cost center -- not a hypothetical -- for any company sitting on large volumes of user-generated content that touches copyrighted material. For enterprise buyers and investors, a credible sanctions finding would add real tail risk to OpenAI's litigation exposure at a moment when the company is simultaneously trying to project stability around its Microsoft relationship and its next funding round.
The bear case for the plaintiffs: OpenAI's fair-use defense doesn't depend on the discovery dispute being resolved in its favor, and courts frequently decline the most aggressive sanctions requested even when some misconduct is found. What to watch next: whether the judge rules on the sanctions motion before the case's next substantive hearing, and whether other publishers suing OpenAI move to join the sanctions request or file parallel motions citing the same Project Giraffe disclosures.
