NYT Accuses OpenAI of Hiding Evidence in Copyright Lawsuit

The New York Times and The Daily News have accused OpenAI of lying about its ability to search chat log data and training datasets for their copyrighted works. The allegation, filed as a new motion for sanctions, is the latest chapter in a two-year lawsuit in which the outlets claim the company violated copyright law by training its models on their journalism and reproducing it in outputs to users.

The dispute matters because it touches the most sensitive nerve in the generative AI industry: what material goes into the models, whether that material is protected, and whether a chatbot's outputs reproduce — or "regurgitate," in the case's parlance — someone else's text. What's now at stake isn't just whether that happened, but whether OpenAI deliberately hid the tools that could prove it.

What the outlets allege

Throughout the case, OpenAI has argued that it lacked the ability to search its own training corpus. It also argued that searching or producing its massive collection of ChatGPT conversations would be technically burdensome and would raise user-privacy concerns, since the logs would need to be retrieved, processed, and de-identified.

The plaintiffs wanted exactly that data: they sought to determine whether their copyrighted journalism was present in OpenAI's training dataset and whether — and how often — ChatGPT generated outputs using or reproducing their content.

The turn, according to the lawsuit, came with an April court-ordered deposition. In it, OpenAI data privacy engineer Vinnie Monaco allegedly revealed that the company had already conducted internal searches and evaluations of its training corpus to search for copyrighted journalism works. In other words, what it claimed it couldn't do, it had already done.

A database and a filter that change everything

That same deposition allegedly exposed two particularly significant elements.

The first: even before the Times filed its lawsuit, OpenAI had already amassed a database of about 78 million de-identified ChatGPT conversations that it was using internally to calculate how much it was infringing on others' works.

The second: shortly after the lawsuit was filed, the company allegedly implemented a filter called "Bloom," part of a set of tools called "Project Giraffe," which detected and kept a record of regurgitation instances in the model's outputs.

Put another way, the plaintiffs accuse OpenAI of having had, this whole time, specific tools to measure the problem, while telling the court that searching for that information was infeasible.

The fight over chat log samples

The other front of the conflict is the production of chat logs. The plaintiffs originally requested a sample of 120 million conversations. OpenAI negotiated it down to 20 million, and delivered that sample to the courts this past December.

The problem, according to the outlets, is that the sample arrived with so many redactions that the court itself called it "unusable." The plaintiffs go further, claiming OpenAI deleted billions of ChatGPT outputs after the lawsuit was filed, in direct violation of the court's preservation order, and that the company substituted millions of logs within the requested sample.

The lawyers' reading is direct: OpenAI allegedly made it needlessly difficult to obtain information the company had already collected on its own.

"If OpenAI genuinely believed that copying our clients' journalism was fair and legal, it wouldn't have hid the truth about having done it," said Ian B. Crosby, lead counsel for the plaintiffs, in a statement.

What they're asking the judge

The NYT and The Daily News are now asking the judge to sanction OpenAI for allegedly withholding evidence and interfering with the discovery process. Specifically, they are asking the court to:

• Prevent OpenAI from using the 20-million chat log sample as evidence, on the grounds that it is unreliable.
• Accept as established fact that ChatGPT logs would have shown significant regurgitation and grounding of the plaintiffs' content in outputs.
• Prevent OpenAI from arguing that the logs it provided don't demonstrate substantial regurgitation.
• Order OpenAI to pay the legal fees incurred in having to chase down this evidence.

These requests are not minor. If the judge grants them, he would effectively be treating as proven facts that would normally need to be established through the very evidence now in dispute. A sanction of this kind can tilt a case even before it reaches the merits.

OpenAI's response

OpenAI spokesperson Drew Pusateri denied the allegations and accused the Times of trying to access private user conversations as its case weakens.

"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," Pusateri said. "We'll continue defending our users' privacy and the long-established principles of fair use."

OpenAI's defense rests, as in other generative AI lawsuits, on the doctrine of fair use — the figure in U.S. law that allows reuse of protected material under certain conditions without the rights holder's permission. But the current dispute isn't about that doctrine; it's about something more fundamental: the company's conduct during the litigation process itself.

Why this standoff matters beyond the case

The clash between news publishers and AI companies has long been defining one of the industry's biggest legal frontiers. Language models are trained on massive amounts of text, and a good portion of that material comes from outlets that never gave permission or received compensation. The underlying question — whether that's transformative training or large-scale copying — still has no firm answer in the courts.

What makes this episode different is that the conflict has shifted from questions of principle to questions of procedural conduct. It's no longer just a debate over whether training on copyrighted journalism is legal, but whether OpenAI hid the tools that could answer that question. For a company that publicly presents itself as a champion of transparency and well-grounded fair use, the accusation of having hidden evidence is doubly uncomfortable.

There's also a tension running through the entire litigation: user privacy versus the plaintiffs' right to obtain evidence. OpenAI maintains that handing over chat logs exposes data belonging to people with no connection to the case; the outlets counter that the company already had de-identified versions of that data and was using them internally. Both things can be true at once, and that ambiguity is exactly what the judge will have to resolve.

The next move is up to the court, which must decide whether to impose the requested sanctions. Its ruling won't just affect the standoff between the Times and OpenAI — it will help define how much AI companies can conceal about their own systems when the law requires them to show their hand.