The news
In a discovery order filed on October 19, 2026, US Magistrate Judge Thomas Hixson ordered Meta to name every person who made, authorized or approved the decisions to obtain books from shadow libraries such as LibGen and Anna's Archive for training its Llama AI models — including whoever decided whether Meta's computers would seed those files to strangers over BitTorrent. The order covers the copyright cases against Meta in San Francisco, including the class action by authors such as Richard Kadrey and Sarah Silverman, and a separate suit from Entrepreneur Media, publisher of Entrepreneur magazine. It is a discovery ruling, not a finding that Meta broke the law: it settles what Meta must disclose.
Background and context
The dispute centers on how Meta assembled the training data for its first two Llama models. The company obtained books from shadow libraries — pirate repositories of copyrighted works — including via BitTorrent. BitTorrent's protocol shares pieces of a file with other users while downloading, so torrenting a pirate library can amount to distributing copies to strangers. That distinction is crucial: in June 2025, US District Judge Vince Chhabria ruled that Meta's training on the authors' books was fair use on the arguments the authors made — leaving the torrenting and distribution claims alive. In copyright law, unauthorized distribution is a separate wrong from unauthorized copying, and it is the distribution claims that now carry the litigation's momentum.
Meta's answer to interrogatories — written questions asking it to identify "each natural person who made, authorized, or approved any decision to obtain or download" a shadow-library dataset — was that no one had been identified who "formally" made such decisions. In an October 6 filing, Meta said decisions were "often collaborative or the result of team consensus." The publisher's lawyers skewered that: "On that construction no one at Meta decided to torrent anything, yet Meta torrented shadow libraries repeatedly from 2022-2024."
What it means / key details
- The rejection of "team consensus": Hixson wrote that Meta's response "wanders away" from the question, and held that "if a decision is fairly attributable to a group or a team, everyone in the group or team who participated in making that decision" must be named, with rough dates and who decided what.
- Scope of disclosure: Meta must name who approved obtaining LibGen data within FAIR, Meta's AI research lab, and who approved resuming a download from Anna's Archive — or state under oath that it has found nobody.
- Seeding under the microscope: Meta must name everyone in the discussion before the engineer's LibGen torrent download, everyone who approved going ahead by torrent, whoever directed or approved an April 2024 download, and anyone who decided "whether or how to seed, or to mitigate seeding" when pulling LibGen's Scimag collection and Anna's Archive content.
- Why the names matter: Entrepreneur Media says the identities bear on whether any infringement was willful, which can raise damages. Internal evidence already surfaced includes a log one employee read before downloading from Anna's Archive: "Decided to go with direct file uploads without using torrents for the following reasons: Using torrents would entail seeding the file such as sharing content outside. This would be legally not okay."
- Collateral disclosures: The order also revealed that Meta licenses multimedia content from Fox Corp. to train AI (Meta accidentally produced unredacted copies of the deal, which Hixson ordered destroyed), and requires Meta to answer questions about stripping copyright management information — author names and notices — rather than dismissing it as "repetitive text," plus what it did with books it downloaded but never used for training.
Why it matters for founders
The order is a warning about data provenance. Courts are increasingly treating how AI training data was gathered — not just how it was used — as fair game for litigation. Founders building AI products should assume that every dataset's chain of custody will eventually be examined: who chose it, who approved it, what alternatives were rejected and what internal concerns were raised. Document decisions in writing, avoid shadow libraries entirely, and assume internal concerns will surface in discovery. The companies surviving this era will be the ones that built licensing and provenance hygiene early, not the ones that optimized for training-data scale first.
Why it matters for investors
For investors in AI, the case reframes copyright risk: the industry's early litigation has focused on whether training is fair use, and courts are beginning to answer that question — but distribution claims, willfulness findings and statutory damages can attach to conduct that fair use doesn't cover. Meta's accidental disclosure that it licenses content from Fox Corp. hints at where the market is heading: licensed data as the compliance norm, with licensing budgets becoming a real line item in model training. Due diligence on AI startups should now include a hard question about training-data provenance — where the data came from, who can document the chain, and whether any ingestion channel resembles what Meta is being forced to explain.
Angeled.com connects founders with investors.
Read Next
Open Weights Are Not Open Business Models: Mistral Large 4 and the Licensing Trap
Courts Are Writing AI's Rules Now: Three Rulings Founders Should Read
Delhi High Court Orders 24-Hour Takedown of AI Deepfake Videos Targeting Chief Minister
EU Court Adviser Says AI Training Is Legal — but Gemini's Answers May Still Infringe Copyright
AI Regulation After the Midterms: What Founders Should Do Before the January Bills Arrive
DOJ Launches Super Intelligence Task Force to Police AI Misuse



