Here’s a legal risk every AI company should be watching closely. The New York Times and The Daily News just accused OpenAI of lying to a federal court about what it can search inside its own systems. If a judge agrees, OpenAI could lose the ability to defend itself with its own evidence, and the fallout could reshape how every AI lab handles discovery in copyright fights.
This is the latest escalation in a two-year lawsuit accusing OpenAI of training its models on the Times’ journalism and reproducing that work in ChatGPT outputs. What stands out here isn’t the copyright claim itself. It’s the accusation that OpenAI had answers the whole time and buried them.
What the plaintiffs claim
Throughout the case, OpenAI argued it couldn’t search its own training corpus. It also said digging through its massive pile of ChatGPT conversations would be technically painful and a privacy nightmare, since logs would need to be retrieved, processed, and de-identified.
Then came an April deposition. OpenAI data privacy engineer Vinnie Monaco allegedly revealed a very different picture:
- OpenAI had already run internal searches of its training data hunting for copyrighted journalism.
- Before the Times even filed suit, OpenAI had built a database of roughly 78 million de-identified ChatGPT conversations to measure how much it was infringing on others’ work.
- Shortly after the lawsuit landed, OpenAI added a “Bloom” filter under a toolset called “Project Giraffe” that detected and logged when outputs regurgitated protected content.
In plain terms: the plaintiffs say OpenAI already had the tools and the data, then told the court it didn’t.
Why the sample matters
The discovery fight got uglier from there. Plaintiffs originally asked for 120 million chat logs. OpenAI negotiated that down to 20 million, then handed over a sample last December so heavily redacted the court itself called it “unusable.”
The accusations go further. Plaintiffs say OpenAI deleted billions of ChatGPT outputs after the suit was filed, violating a court preservation order, and swapped millions of logs in the requested sample. Their lead counsel, Ian B. Crosby, put it bluntly: “If OpenAI genuinely believed that copying our clients’ journalism was fair and legal, it wouldn’t have hid the truth about having done it.”
What they want the judge to do
The Times and The Daily News are now asking the court to punish OpenAI. Specifically, they want the judge to:
- Bar OpenAI from using the 20 million log sample as evidence, calling it unreliable.
- Accept as established fact that the logs would have shown major regurgitation of the plaintiffs’ content.
- Block OpenAI from arguing its logs don’t prove substantial copying.
- Make OpenAI pay the legal fees spent chasing this evidence.
Those are called sanctions, and they’re serious. When a court accepts a fact as proven because one side hid evidence, the case can tilt hard before it ever reaches a verdict.
OpenAI flatly denies all of it. Spokesperson Drew Pusateri accused the Times of trying to crack open private user chats as its case falls apart. “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,” he said, adding the company will keep defending “our users’ privacy and the long-established principles of fair use.”
Why this matters for the industry
The legal question of whether training on copyrighted work counts as fair use is still open. But this fight has quietly shifted to something more dangerous for AI companies: conduct during discovery. Courts have far less patience for hiding evidence than for aggressive legal theories.
If you build or run AI products, treat this as a warning. “We can’t search our own data” is getting harder to say with a straight face, and preservation orders mean what they say. The companies that document their data honestly now will be in a much stronger spot when their own subpoena arrives.
A ruling on the sanctions request will be the next thing to watch. You can find the full details at the original TechCrunch AI report.