NYT v. OpenAI: The Evidence Publishers Accuse the Company of Hiding in the Copyright Case

Two and a half years after The New York Times sued OpenAI and Microsoft over the use of its articles to train ChatGPT, the lawsuit has entered new territory: the fight is no longer just about whether copyright was infringed, but about whether one side has played fair with the evidence. On July 9, the Times and other major U.S. publishers — including the Daily News, the Chicago Tribune, the Center for Investigative Reporting, The Intercept and Ziff Davis — asked the federal court in the Southern District of New York to sanction OpenAI for withholding and destroying evidence (a motion docketed as document 1618 in the consolidated case, 1428 on the Times docket). The motion was provisionally filed under seal because it contains protected material, but its 52-page memorandum is public, with partial redactions.

What the publishers say was hidden

The memorandum opens bluntly: "This is a case about copying. There is no question that it happened. (…) 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." The core accusation is a documented contradiction between what OpenAI told the court and what its own witness later acknowledged.

In September 2024, according to the memorandum, OpenAI told the court it did "not currently have tools to search [the training datasets] efficiently"; in May 2025, that searching ChatGPT's output logs for specific content would require designing "a custom system." But in April 2026, in a second deposition the court ordered precisely because the witness had been unprepared the first time, data privacy engineer Vincent Monaco acknowledged — according to the plaintiffs — that the company had been running exactly those searches for years: an internal program called Project Giraffe tracked "regurgitations" (near-verbatim reproductions of third-party text in the chatbot's answers), and OpenAI kept a set of 78 million de-identified ChatGPT conversations it used internally to measure the phenomenon.

There is more: the publishers contend that OpenAI deleted "billions" of output logs, even after a court order to preserve them, and that the sample of 20 million conversations it did produce arrived with redactions that render it "unusable." Both, it bears underlining, are one side's characterizations: OpenAI has not yet filed its formal opposition.

Spoliation, in plain language

In U.S. civil procedure, spoliation means destroying, altering or letting perish evidence a party had a duty to preserve for ongoing or reasonably foreseeable litigation. Federal Rule 37 gives the judge a menu of responses: from deeming certain facts established to the so-called adverse inference — instructing the jury to presume that what was destroyed was unfavorable to the party that destroyed it — reserved for cases where an intent to deprive the other side of the evidence is shown. The essential point for the reader: a sanctions motion does not decide who is right about copyright. It decides which evidence, and under which presumptions, that game will be played.

What exactly they are asking for, and OpenAI's answer

The publishers request five measures: that OpenAI be barred from relying on the 20-million-conversation sample "for any purpose"; that the court deem it established that ChatGPT's logs show (or would have shown) "substantial and systematic" grounding on and regurgitation of their content; that OpenAI be barred from arguing otherwise; that the jury be instructed on these findings with binding effect; and an award of attorneys' and expert fees.

OpenAI, through spokesperson Drew Pusateri, rejects the motion outright: "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."

The immediate clock is already running: the defendants have five business days to justify which parts of the motion should remain sealed, and then come OpenAI's opposition and a decision by Magistrate Judge Ona T. Wang, who oversees discovery under Judge Sidney H. Stein. No date is set on the docket.

The Times' original complaint was filed on December 27, 2023, and is now part of consolidated litigation grouping publishers' and authors' cases against OpenAI. If the motion succeeds on its harshest terms, the jury would arrive at trial with a weighty presumption against the company; if it fails, it will stand as one more episode in an already long procedural war. In a case poised to define how copyright applies to the generative AI era, the battle over the evidence is not a prelude: it is very nearly the whole game.