Meta secretly trained its AI on a notorious hacking database, newly redacted court documents reveal.

“Meta has treated the so-called “general availability” of parallel data sets as a get-out-of-jail-free card, even though internal Meta records show every relevant decision-maker at Meta, including its CEO, Mark Zuckerberg, was Knows that LibGen was “a data set that we know was pirated,” the plaintiffs allege in this suit (originally filed in late 2024, and is a request for a third amended complaint).

In addition to plaintiffs’ briefs, another unredacted motion was filed in response to Chhabria’s order—the Meta Motion opposition Regarding the request to submit an amended complaint. She argues that the authors’ attempts to add additional allegations to the case are “an eleventh-hour maneuver based on a false and inflammatory premise” and denies that Meta waited to reveal important information in discovery. Instead, Meta says it first disclosed to prosecutors that it used the LibGen dataset in July 2024. (Because much of the material discovered remains classified, it is difficult for WIRED to confirm this claim.)

Meta’s argument relies on its claim that the plaintiffs were already aware of the use of LibGen and should not have been given additional time to file a third amended claim when they had ample time to do so before discovery expired in December 2024. And the use of LibGen and other alleged “shadow libraries” since mid-July 2024 at least. He argues.

In November 2023, Chhabria granted Meta’s request to dismiss some of the lawsuit’s claims, including its claim that Meta’s alleged use of authors’ work to train artificial intelligence violates rules of international law. DMCAan American law introduced in 1998 to prevent people from selling or copying copyrighted works on the Internet. At that time the judge Agreed With Meta’s position that the plaintiffs did not provide sufficient evidence to prove that the company removed what is known as “copyright management information,” such as the author’s name and title of the work.

The unredacted documents say the plaintiffs should be allowed to amend their complaint, claiming that the information revealed by Meta is evidence that their DMCA claim was justified. They also say the discovery process has revealed reasons to add new claims. The lawsuit alleges that “Meta, through a company representative who testified on November 20, 2024, has now admitted under oath to uploading pirated files (also known as “seeding”) containing Plaintiffs’ works on torrent sites.” (Seeding occurs when files are shared Torrent with other peers after the download is finished.)

“This torrenting activity resulted in Meta turning itself into a distributor of the same pirated copyrighted material that it was also downloading for use in commercially available AI models,” one newly redacted document claims, claiming that Meta, in other words, did not So you have just used copyrighted material without permission but also posted it.

LibGen, an archive of books uploaded to the Internet that originated in Russia around 2008, is one of the largest and most controversial “shadow libraries” in the world. In 2015, New York Judge commander A preliminary injunction was ordered against the site, an action theoretically designed to temporarily shut down the archive, but its anonymous administrators simply changed its domain. In September 2024, another New York judge commander LibGen will pay $30 million to rights holders for violating their copyrights, although it is not known who actually runs the piracy center.

The problems with meta-detection of this case are not over either. In the same vein, Chhabria warned the tech giant against any overly broad redaction requests in the future: “If Meta again makes an unreasonably broad shutdown request, all materials will simply be unlocked,” he wrote.