A 16-member coalition led by The New York Times says OpenAI concealed its ability to search training data and may have destroyed 20 million ChatGPT logs, escalating a nearly three-year-old copyright fight now watched as a template for AI licensing.
Discovery fights rarely make headlines on their own, but in the sprawling litigation over how AI companies train large language models, what a defendant did or didn’t hand over during discovery has become the case. On July 9, a coalition of roughly 16 news organizations led by The New York Times asked a federal court in Manhattan to sanction OpenAI, arguing that the company spent more than two years misrepresenting what it could search for and may have destroyed evidence central to proving infringement, according to Crypto Briefing.
the core allegation
The publishers’ filing, obtained and viewed by TheWrap and reported by Yahoo News Canada, doesn’t mince words: “This is a case about copying. There is no question that it happened,” the filing states. It continues: “It told News Plaintiffs and the Court, that it could not search for evidence that it copied their works to train its models, or again to ground their outputs, or yet again in the outputs themselves. None of this was true. OpenAI had done all these things — starting even before the first News Plaintiff filed suit.” The coalition, which includes the New York Daily News and Chicago Tribune alongside the Times, says OpenAI concealed its ability to search training datasets for over two years, per Crypto Briefing. The publishers originally sought a sample of 120 million chat logs; OpenAI negotiated that down to 20 million and allegedly substituted millions of logs within that smaller set, according to bestmediainfo.com.
the deposition problem
The most damaging material, by the publishers’ account, surfaced in an April deposition. OpenAI data privacy engineer Vinnie Monaco reportedly said the company had already run internal searches of its training data for copyrighted journalism — something OpenAI had previously told the court and plaintiffs it could not do, bestmediainfo.com reported. The same deposition allegedly revealed an internal database of roughly 78 million de-identified ChatGPT conversations, built to gauge infringement frequency, plus a tool called a “Bloom filter” under an internal project named “Project Giraffe” that tracked regurgitation in outputs. New York Times lead attorney Ian Crosby, as cited by Storyboard18, said OpenAI had misled the newspapers, the public and the court for more than two years by claiming searches of ChatGPT outputs for copies of the publishers’ content were not feasible while allegedly conducting those searches internally. Daily News attorney Steven Lieberman told Reuters, as relayed by bestmediainfo.com, that the motion is meant to “punish OpenAI for hiding and destroying evidence showing how ChatGPT was trained on stolen journalism.”
what OpenAI says
OpenAI disputes the characterization. A spokesperson told media, per Yahoo News Canada: “As the Times’ case weakens and they’ve been forced to drop claims against us, they’re persisting with their efforts to invade the privacy of people who have nothing to do with this case, including by making these blatantly false allegations. We’ll continue defending our users’ privacy and the long-established principles of fair use.” Spokesperson Drew Pusateri offered a similar line to Reuters, according to bestmediainfo.com, framing the sanctions push as retaliation for a weakening case rather than genuine spoliation. Notably, Microsoft — named as a co-defendant in the original 2023 suit for allegedly using millions of the Times’ articles to train the models behind ChatGPT — is not a target of this particular sanctions filing, Crypto Briefing noted.
why it matters for the industry
The stakes go beyond one lawsuit. The Times sued OpenAI and Microsoft in December 2023, and the case was later consolidated with suits from other publishers, including the Daily News and its seven sister papers, bestmediainfo.com reported. Both defendants’ motions to dismiss were denied on the core copyright claims, which are now in discovery — precisely the stage where the publishers say OpenAI stonewalled, per Crypto Briefing. Sanctions motions like this one ask judges to consider an adverse-inference instruction: if a court finds OpenAI failed to preserve evidence, jurors could be told to assume the missing logs would have supported the infringement claims, a presumption Crypto Briefing called potentially “devastating for OpenAI’s defense.” The dispute sits alongside a wider wave of AI training-data litigation — a December suit from journalists including former Times reporter John Carreyrou and ex-Wall Street Journal reporter Philip Shishkin accused OpenAI, Anthropic, Google, Meta, xAI and Perplexity of illegally copying copyrighted books to train commercial models, per Yahoo News Canada. According to bestmediainfo.com, the Times has reportedly spent more than $28 million on litigation against AI companies, including a separate suit against Perplexity, underscoring how costly and protracted this fight has become for publishers with far shallower pockets. How the judge rules on sanctions — and whether an adverse-inference instruction materializes — will likely shape not just this case’s trajectory but the leverage publishers bring to licensing talks with every major model builder watching from the sidelines.
"This is a case about copying. There is no question that it happened. Nor should there be one about what was copied, how often, or to what end. The evidence is in OpenAI's training datasets and ChatGPT output logs. But instead of just producing that evidence at the start of the case and focusing on the merits of its fair use defense, OpenAI chose obstruction."
