Meta Secretly Trained Its AI on a Notorious Piracy Database, Newly Unredacted Court Docs Reveal

“Meta has treated the so-called ‘public availability’ of the Shadow dataset as a get-out-of-jail-free card, even though internal Meta records show every relevant decision-maker at Meta, up to and including its CEO Mark Zuckerberg, knew that LibGen was ‘a dataset that we know to be pirated. ,’” the plaintiffs alleged in the motion. (Originally filed in late 2024, the motion requested the filing of a third amended complaint.)

In addition to plaintiffs’ briefs, another filing was not redacted in response to Chhabria’s order—Meta’s the opposition On the motion to file an amended complaint. It argued that the authors’ attempt to add additional claims to the lawsuit was an “eleventh-hour gambit based on false and inflammatory premises” and denied that Meta had waited to reveal important discovery information. Instead, Meta argues that it first disclosed to the plaintiffs that it used a LibGen dataset in July 2024. (Because much of the discovery material remains confidential, it’s difficult for WIRED to confirm this claim.)

Mater’s argument rests on its claim that plaintiffs already knew about LibGen’s use and should not be given additional time to assert a Third Amended claim when they had ample time to do so before discovery ended in December 2024. “Plaintiffs knew about downloading Meta. and use of LibGen and other so-called ‘shadow libraries’ from at least mid-July 2024,” the tech giant’s lawyer to argue.

In November 2023, Chhabria granted Meta’s motion to dismiss some of the lawsuit’s claims, including claims that Meta infringed on its use of the authors’ work to train AI. Digital Millennium Copyright ActA US law was introduced in 1998 to prevent people from selling or copying copyrighted works on the Internet. At this time, the judge agreed Mater’s position is that the plaintiffs did not provide sufficient evidence to prove that the company removed what is known as “copyright management information” such as author names and work titles.

The unredacted documents argue that the plaintiffs should be allowed to amend their complaint, alleging that the information that Meta disclosed proved that the DMCA claims were true. They also say that the discovery process found grounds to add new charges. “Meta, through a corporate representative who testified on November 20, 2024, now admits under oath to uploading (aka ‘seeding’) pirated files including Plaintiff’s work to ‘torrent’ sites,” the motion alleges. (Seeding is when torrented files are shared with other peers after they have finished downloading.)

“This torrenting activity made Meta a distributor of the same pirated copyrighted material it was downloading for use in its commercially available AI models,” a new unredacted document claims. It has been promoted.

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, a New York judge order An initial ban against the site, a measure theoretically designed to temporarily shut down the archive, but its anonymous administrators simply changed its domain. In September 2024, a different New York judge order LibGen will pay rights holders $30 million for copyright infringement, despite not knowing who actually operates the piracy hub.

In this case, the problem of the discovery of matter is not over. In the same vein, Chhabria warned the tech giant against any overly broad reformation requests in the future: “If Meta again submits an unreasonably broad unsealing request, all materials will simply be unsealed,” he wrote.