On October 20, federal magistrate judge Thomas Hixson ordered Meta to identify every person who 'accepted, authorized, or approved' decisions to download books via BitTorrent for training artificial intelligence (AI) models. The order goes beyond the employees who technically performed the downloads: the court is asking who made the decisions — including who decided to seed the books to others across the torrent network. It is the latest turn in the ongoing copyright litigation against AI companies.
What the Order Covers
The order names two shadow libraries by name — Anna's Archive and LibGen. Publisher Entrepreneur Media had asked the court to compel Meta to identify everyone who accepted, authorized, or approved decisions to obtain these 'shadow library datasets' via BitTorrent. Judge Hixson granted the request. The order is not limited to a list of names: the company must also explain the circumstances under which the decisions were made and answer questions about the fate of pirated copies that were not used in AI training. In other words, the court is looking not only for 'who downloaded,' but also for 'who authorized it' and 'where the remaining copies are.'
"A federal magistrate judge ordered Meta to identify every person who 'accepted, authorized, or approved' decisions to obtain books from shadow libraries via BitTorrent — going beyond the employees who technically performed the downloads and also demanding to know who decided to seed the books to others over the torrent network." — Ars Technica
The 'Collective Decision' Defense Failed
In its defense, Meta said that obtaining books for its first two Llama models was never 'formally' approved by anyone. The company's position is that decisions were made by collective agreement, with no single decision-maker — so there is no one to name. Judge Hixson rejected that narrow reading. Under his order, Meta must answer every one of the following questions: who made the decision, who authorized it, who ratified it, who directed it, who permitted it, or who approved it — and under what circumstances it happened. The formulation effectively nullifies the company's 'nobody formally approved it' defense: the court is asking not for an official signature, but for the actual chain of decision-making. The history of Meta's internal decisions may thus be disclosed in an official court document for the first time.
Four New Lawsuits and a $1.5 Billion Precedent
A day earlier, on October 19, four related new copyright lawsuits were filed against Meta in the U.S. District Court for the Northern District of California: Reno v. Meta (3:26-cv-12131), Sheard v. Meta (3:26-cv-12126), Winspear v. Meta (3:26-cv-12100), and Older v. Meta (3:26-cv-12115). According to Bloomberg Law, the suits — brought by hundreds of authors — accuse the company of unlawfully training AI models on protected works. The cases are related to the earliest one, Kadrey v. Meta. In Kadrey, the court found training models on books to be fair use, leaving the BitTorrent seeding claims as the case's only remaining live part. Another heavy precedent hangs over the proceedings: Anthropic's $1.5 billion piracy settlement in the Bartz case. The sum is being assessed as one of the largest settlements ever paid in copyright disputes against AI companies, and it raises the financial stakes of Meta's cases.
Why Seeding Is the Decisive Issue
At the heart of the legal dispute is the distinction between two acts: downloading books via torrent and distributing them to others through the same network. By the nature of the BitTorrent protocol, a downloader simultaneously becomes a seeder — the computer receiving the file automatically relays its pieces to other participants. Although the court recognized AI training on books as fair use, that protection was not extended to acts of distribution. That is Meta's most vulnerable point: even if the download itself is excused, re-sharing thousands of books across the torrent network is being treated as a separate violation. Judge Hixson's order rests on the same logic — beyond the employees who ran the downloads, those who decided to distribute the books to others must also be identified. The company is now compelled to answer these questions, and the answers could shape future copyright case law in the AI industry.
The order was issued on October 20 — the day after the four new lawsuits were filed. The suits and the order reinforce each other: while the new suits challenge the legality of the training, the judge's order targets the people responsible in the distribution chain. Both lines of inquiry seek an answer to one question — where the boundary lies in the use of copyrighted works to build AI models.
