A federal judge ruled that OpenAI needs to turn over all its internal communications with lawyers about why it deleted two massive troves of pirated books from a notorious “shadow library” that the tech company is accused of using to train ChatGPT.

Manhattan Federal Court Magistrate Judge Ona Wang ruled Monday that the tech giant’s shifting reasons for deleting the data tanked any argument that those reasons could be protected by attorney-client privilege.

“OpenAI continues to assert that it did not willfully infringe Class Plaintiffs’ copyrighted works. A jury is entitled to know the basis for OpenAI’s purported good faith,” Wang wrote in her 28-page decision. “What matters is that OpenAI has put its state of mind at issue, and OpenAI may not selectively use attorney-client privilege to restrict Class Plaintiffs’ inquiry into evidence concerning OpenAI’s purported good faith in this way.”

The judge is overseeing a massive consolidated class-action lawsuit against Microsoft and  OpenAI, which includes the Daily News, affiliated newspapers at Tribune Publishing and MediaNews Group and other news outlets that are accusing the technology giant of copyright infringement.

Wang’s decision Monday centers on a group of plaintiffs that include the Authors Guild and a long list of best-selling writers like “A Game of Thrones” scribe George R.R. Martin and legal thriller author John Grisham. The authors allege that OpenAI used pirated books from the infamous online “LibGen” library, which two courts have ordered shut down over the past decade, to train its AI products, after an employee downloaded them in 2018.

During the discovery process, the plaintiffs found out that OpenAI deleted the two troves, called “Books1” and “Books2,” in 2022 — believed to contain more than 100,000 books — a year before any litigation began.

“At the time, OpenAI asserted that the datasets were deleted due to ‘non-use.’ These are the only training datasets that, according to OpenAI, have ever been deleted,” Wang wrote. “Then, when Class Plaintiffs sought discovery about the reasons for the deletion of the Books1 and Books2 datasets, OpenAI asserted attorney-client privilege. OpenAI’s position on whether the reasons for the deletion are privileged has shifted several times.”

Wang is ordering OpenAI to give the plaintiffs communications she’s already reviewed, all other written communications with the company’s in-house lawyers regarding the reasons the datasets were deleted, and all internal references to LibGen that OpenAI has previously redacted or withheld.

The Authors Guild and OpenAI’s legal teams did not immediately return messages seeking comment.

An OpenAI spokesperson told Law360, “We disagree with the ruling and intend to appeal.”