Meta Secretly Trained Its AI on Well-Known Piracy Network, New Unedited Court Docs Reveal

“Meta has taken the so-called ‘public access’ of the screenshots as a free card, even though Meta’s internal documents show everyone involved in decision-making at Meta, including its CEO, Mark. Zuckerberg, knew that LibGen was ‘information we know to be pirated,’ ” the plaintiffs say in this case. (Originally filed in late 2024, this petition is a third amended complaint.)

In addition to the plaintiffs’ brief, other reservations were not made pursuant to Chhabria-Meta’s order. opposition to a request to file an amended complaint. It argues that the authors’ efforts to add to the allegations in the lawsuit are an “eleventh-hour gambit based on falsehoods and inflammatory statements” and denies that Meta waited to reveal the necessary findings. Instead, Meta contends that it first disclosed to the plaintiffs that it had used LibGen data in July 2024. (Because many of the findings remain confidential, it is difficult for WIRED to verify their claims.)

Meta’s argument is based on the claim that the plaintiffs already knew about the use of LibGen and should not be given an extension to file a third amended motion after they had enough time to do so before December 2024. and the use of LibGen and other so-called ‘image libraries’ from mid-July 2024,” the tech giant said. lawyers to argue.

In November 2023, Chhabria accepted Meta’s request to dismiss the lawsuit, including claims that Meta’s use of the authors to train AI violated the law. Digital Millennium Copyright Acta United States law enacted in 1998 that prohibits people from selling or copying illegal works on the Internet. At that time, the judge he agreed It is Meta’s opinion that the plaintiffs did not provide sufficient evidence to prove that the company removed what is known as “copyright control information,” such as the name of the author and the title of the work.

The amended filings argue that the plaintiffs should be allowed to amend their complaints, arguing that Meta’s information was disclosed by evidence that the DMCA claimed was appropriate. They also said that this method of discovery has discovered reasons for adding new reasons. “Meta, through a company representative who testified on November 20, 2024, agreed to download (aka ‘seed’) fraudulent files containing the Plaintiffs’ services on ‘torrent’ sites,” it says. (Seeding is when viewers of streams then share with their friends after they’ve finished downloading.)

“This operation turned Meta itself into a distributor of copyrighted material that was re-downloaded for use in its commercially available AI models,” one of the unredacted notes said, stating that Meta, in other words, simply used the licensed material without permission and distributed it. .

LibGen, an online library that originated in Russia around 2008, is one of the largest “image libraries” in the world. In 2015, a New York judge he ordered the original law against the site, a measure designed to temporarily close the storage facility, but its anonymous administrators simply changed its location. In September 2024, a New York judge he ordered LibGen to pay $30 million to copyright holders for copyright infringement, although they did not know who was responsible for the crime.

The problems encountered by Meta in this case are not over yet. Likewise, Chhabria warned the chief engineer not to make any further requests in the future: “If Meta makes another request for an unwarranted publication, all materials will be automatically released,” he wrote.