Meta Must Name Who Approved Shadow Library Downloads

A US magistrate judge has ordered Meta to identify every person who made, authorised or approved decisions to obtain or download pirated “shadow library” datasets used in building its AI models. The discovery order, signed by Magistrate Judge Thomas S. Hixson in San Francisco on 19 October, rejects Meta’s argument that early decisions were informal team calls with no one in charge.

The ruling matters because Entrepreneur Media, one of the companies suing Meta, says the answers “bear on” wilfulness, TorrentFreak reported. It is a procedural step about evidence, not a finding that Meta broke the law. TorrentFreak first reported the order, and Ars Technica followed.

In brief

A magistrate judge has ordered Meta to name everyone who made, authorised or approved decisions to obtain shadow libraries such as LibGen and Anna’s Archive.
The order rejects Meta’s argument that there was no single decision maker; everyone in a deciding team must be named.
It is a discovery order only: no court has found Meta liable for the BitTorrent downloads.

What the judge decided

The dispute came from Entrepreneur Media, the company behind Entrepreneur magazine, which asked Meta to “IDENTIFY each natural person who made, authorized, or approved any decision to obtain or download any Shadow Library Dataset containing Plaintiff’s Works.” Hixson granted that request in full.

He found Meta’s answers drifted off the question. For its first two Llama models, Meta said it had not identified anyone who “formally” made or approved decisions to acquire or use books. “The rog isn’t limited to formal decisions,” the judge wrote, using legal shorthand for an interrogatory, adding that the request is about obtaining and downloading, not use. “This is the question Meta must answer.”

Meta’s team-consensus defence also failed. The request “does not presume that Meta had a single decisionmaker; it asks about all the decisonmakers,” Hixson wrote. “If a decision is fairly attributable to a group or a team, everyone in the group or team who participated in making that decision is responsive to this rog.”

Torrenting and seeding

A second request went further, covering BitTorrent. The judge ordered Meta to identify who took part in the discussion before an engineer torrented LibGen data, who approved going ahead by torrent and who directed an April 2024 download from Anna’s Archive. Meta must also name everyone involved in “the decision whether or how to seed, or to mitigate seeding” when it obtained LibGen’s Scimag collection and Anna’s Archive content by BitTorrent.

Seeding is the heart of the remaining claims, because sharing files back out could count as distribution. Meta has said it “is not currently aware of any persons who made, authorized, or approved any decision to upload or distribute” the datasets, and that it used mitigations to limit seeding, Ars reported. Entrepreneur Media countered: “On that construction no one at Meta decided to torrent anything, yet Meta torrented shadow libraries repeatedly from 2022-2024,” according to TorrentFreak.

The authors argue the scale points to seeding. They say Meta downloaded 80.6 terabytes from LibGen and 35.7 terabytes through Anna’s Archive, figures they contend could not have been reached quickly without sharing, Ars reported. Ars also said unsealed messages allegedly showed the LibGen decision came after “a prior escalation to MZ”, a reference to Mark Zuckerberg. Meta has not been found to have done anything wrong on that point.

A wider batch of rulings

The order is one of several in the related cases against Meta. In a suit led by Chicken Soup for the Soul, Hixson ordered Meta to explain what, if anything, it did with downloaded works that were never used for model training, calling that “relevant and discoverable information”. He also granted Fox Corp.’s request to shield the unredacted terms of its AI content licence with Meta.

The bigger questions remain open. In the authors’ case led by Richard Kadrey and Sarah Silverman, the court has already ruled that training on the books was fair use, leaving the BitTorrent claims as the last live part, TorrentFreak noted. Whether the torrenting itself was lawful, and whether Meta distributed books to others, has yet to be decided. Meta did not immediately respond to Ars’ request for comment.

Our read

This is TSN’s analysis. Courts have so far been kinder to AI training than to how the training data was collected, and this order keeps the focus on acquisition. Making Meta put names to decisions turns a debate about corporate process into one about individual knowledge. The next test is who appears on Meta’s list, and whether it reaches beyond the engineers who ran the downloads.