Short answer: Court records say Meta downloaded collections from LibGen and Anna’s Archive, while later complaints allege that it torrented millions of copyrighted publications and the Books3 collection to obtain training material for Llama. Those figures remain allegations in complaints, not final findings that Meta illegally trained its models. In June 2025, a judge granted Meta a narrow summary-judgment win on the named authors’ copying claim because they had not shown meaningful market dilution; the judge expressly said that decision did not establish that Meta’s training practices were lawful.
Meta, not “Facebook,” is the company in these cases
Facebook’s parent company is Meta Platforms, and the litigation concerns Meta’s Llama language models. The allegations describe obtaining books through shadow libraries and torrent-based networks rather than licensing every title directly from rights holders.
| # | Preview | Product | Price | |
|---|---|---|---|---|
| 1 |
|
Copyright Law | $144.88 | Buy on Amazon |
| 2 |
|
Copyright Law: Cases and Materials (v8.0) | $19.87 | Buy on Amazon |
| 3 |
|
Copyright Law of the United States: and Related Laws Contained in Title 17 of the United States Code | $10.32 | Buy on Amazon |
| 4 |
|
Copyright Law in a Nutshell | $65.00 | Buy on Amazon |
| 5 |
|
Copyright Handbook, The: What Every Writer Needs to Know | $37.99 | Buy on Amazon |
As an Amazon Associate I earn from qualifying purchases.
“Pirated” describes the alleged source of the copies, not a final judicial finding about every file Meta possessed. The procedural distinction matters: complaints contain one side’s allegations, discovery and court orders describe evidence, and rulings decide only particular claims on particular records.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
What the filings say Meta obtained
| Record and date | Collection or transfer | Reported scale | Procedural status |
|---|---|---|---|
| Kadrey v. Meta order, June 25, 2025 | LibGen and Anna’s Archive downloads using BitTorrent | The order does not set a total book count or terabyte figure. | The judge described the downloads in a court order based on the case record. |
| Elsevier complaint, May 5, 2026 | LibGen torrents in 2022 | More than 2,000,000 copyrighted publications | Complaint allegation, not a final finding. |
| Elsevier complaint, May 5, 2026 | Anna’s Archive acquisition | More than 81 terabytes | Complaint allegation, not a final finding. |
| Elsevier complaint, May 5, 2026 | Traffic logs cited for April–July 2024 | 134.6 TB downloaded and 40.42 TB uploaded | Alleged logs described in the complaint; the figures are not a judicial determination that every byte was a copyrighted book. |
| Entrepreneur complaint, November 6, 2025 | Books3, described as derived from the Bibliotik private tracker | 196,640 books | Complaint allegation. The Elsevier complaint separately alleges that Meta later torrented Books3. |
The numbers measure different things. A publication count, a named dataset’s book count, and network traffic in terabytes cannot be added together without knowing the files, duplicates, formats, and other data in each transfer. There is therefore no court-adjudicated total number of books Meta “stole.”
#1 Best Overall
Why BitTorrent matters to the copyright allegations
BitTorrent is designed to exchange pieces of a file among participants. The June 25, 2025 Kadrey order explains that a participant can download pieces while also re-uploading them, a process commonly described as leeching while downloading and seeding after or during completion.
That network behavior creates two legally distinct questions:
Rank #2
- Copying: Did Meta make or keep copies of books for a dataset or model-training process?
- Distribution: Did Meta transmit pieces of those works to other participants through the torrent protocol?
A large upload total may be relevant to the second question, but it does not by itself identify which titles were uploaded, whether a complete work was transmitted, or whether a particular plaintiff’s book was among the files.
Free tools Windows power users keep installed
One-click scans. No signup required.
How the books relate to Llama training
The Kadrey plaintiffs alleged that Meta copied their books to train Llama. The complaints from Elsevier and Entrepreneur similarly portray the shadow-library and Books3 acquisitions as attempts to obtain copyrighted text for artificial-intelligence datasets. The public filings do not establish that every book listed in those collections entered a released model, that every title was used in the same training run, or that Llama memorized and reproduced particular books.
Rank #3
The cases also involve different acquisition methods and purposes. A licensed or public-domain work would have a different authorization status from an allegedly pirated copy; a direct download would present a different distribution question from a torrent that uploads pieces; and using a file for evaluation would not necessarily present the same issue as putting it into training data. The filings cited here focus on alleged acquisition and training-related copying, not a comprehensive inventory of all of Meta’s datasets.
What the June 25, 2025 Kadrey ruling actually decided
The limited win for Meta
Judge Vince Chhabria granted Meta summary judgment on the named authors’ claim that copying their books to train Llama was infringement because the plaintiffs had not produced meaningful evidence that the alleged use diluted the market for their works. Summary judgment means the claim could not proceed on the evidentiary record presented; it was not a ruling after a full trial on every theory or every book.
Rank #4
What the order did not decide
The order did not resolve the alleged distribution claim. It also did not approve all uses of copyrighted books in artificial-intelligence training, decide that every shadow-library file was lawfully obtained, or bind copyright owners who were not the named plaintiffs.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.“This ruling does not stand for the proposition that Meta’s use of copyrighted materials to train its language models is lawful.”
Best Value
Why the March 25, 2026 order matters
On March 25, 2026, the Kadrey court allowed the plaintiffs to add distribution and contributory-infringement theories. The order reiterated that Meta’s earlier summary-judgment victory resulted from the named plaintiffs’ failure to show market harm on the copying theory; it did not create a general safe harbor for AI training on copyrighted works.
Those distribution and contributory-infringement issues remained active after that order. Their eventual outcome cannot be inferred from the earlier copying ruling.
What the licensing allegations add
The May 5, 2026 Elsevier complaint alleges that Meta discussed increasing its dataset-licensing budget from $17 million to $200 million between January and April 2023. It further alleges that licensing efforts stopped after the issue was escalated to Mark Zuckerberg. These are assertions in a filed complaint, not findings that Meta approved either budget or abandoned a legally required licensing program.
The allegation is relevant because it describes an alternative path—paying for rights—alongside the alleged use of shadow-library sources. It does not prove that any particular book was available for licensing, that negotiations covered the books at issue, or that a licensing budget would have resolved the claims.
What readers can safely conclude
- Court records describe Meta downloading from LibGen and Anna’s Archive, and complaints allege very large torrent transfers connected to copyrighted publications and Books3.
- The headline figures—more than two million LibGen publications, more than 81 TB from Anna’s Archive, 196,640 Books3 titles, and the 2024 download/upload totals—come from complaints or cited logs, not final judgments.
- The June 2025 decision was a narrow evidentiary ruling for Meta on the named authors’ copying claim.
- Distribution and contributory-infringement theories were allowed to continue in March 2026.
- No ruling cited here establishes a blanket rule that training an AI model on copyrighted books is lawful or unlawful.
Verdict
It is fair to say that filings allege Meta used torrent-based shadow-library sources, including LibGen, Anna’s Archive, and Books3, while building material associated with Llama. It is not fair to report that a judge found Meta guilty of stealing a settled number of books or declared all of its AI training illegal. The central copyright questions—especially alleged distribution and the legal effect of the training copies—remain narrower and more contested than the headline suggests.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
