Is Anthropic Facing Claims That Claude Was Trained On Pirated Books?

🔍 Read the full analysis: Is Anthropic Facing Claims That Claude Was Trained On Pirated Books? on ThorstenMeyerAI.com

Anthropic is facing new copyright lawsuits from authors who opted out of a class settlement over the company’s downloading of pirated books, but no court has ruled on those new claims. The dispute follows a June 2025 ruling that found training Claude on books was fair use while finding that downloading more than seven million pirated copies to build a permanent library was not.

The 2025 decision came in Bartz v. Anthropic, a case filed by three authors in August 2024. Judge William Alsup said Anthropic’s use of books to train its models was fair use, and also found that the company’s purchase and scanning of print books was fair use. He treated the downloading and retention of pirated copies as a separate matter: the court order described about 196,640 books from Books3, at least five million copies from Library Genesis, or LibGen, and at least two million from the Pirate Library Mirror, or PiLiMi.

The ruling was a pretrial summary-judgment order, not a jury verdict on damages. On the pirated-copy issue, Judge Alsup accepted the authors’ version of the facts; a jury was expected to address remaining questions, but the case settled before trial. The ruling does not mean that a court found every aspect of Anthropic’s training unlawful: it expressly distinguished the act of training from acquiring and retaining pirated books.

A court approved a $1.5 billion settlement in July 2026, covering owners of roughly 500,000 registered books that Anthropic downloaded from LibGen or PiLiMi. The stated payment works out to about $3,000 per book before fees. The settlement requires Anthropic to destroy the downloaded collections and copies, subject to legal-hold obligations, and releases specified claims over past conduct through August 25, 2025. It does not release claims about Claude’s outputs.

What the Ruling Means for Authors

The case draws a legally important line between how an AI model is trained and how the material used to build a training library was obtained. The 2025 ruling found the training at issue to be fair use, but it did not excuse keeping pirated copies. That distinction means the decision should not be read as a blanket permission for AI companies to source copyrighted works from unauthorized collections.

For writers and publishers, the settlement sets a large payment for a defined group of registered books, but it does not resolve every copyright dispute involving AI or establish a general licensing system. The new cases could test claims by authors outside the settlement, while the court has not yet decided whether those allegations succeed. For AI companies, the litigation highlights that data acquisition practices can carry separate legal risks from the later use of material in model training.

From Bartz Ruling to New Suits

The Stanford Daily reported that works by Stanford-affiliated writers, including Tobias Wolff, Abraham Verghese, Chang-rae Lee and Daniel Mason, were among the books said to be affected. That list is the student newspaper’s reporting; the source article says it did not independently verify it against the settlement database. The report also stated that ten of Wolff’s works were affected. Those details should be treated as reported, rather than as findings in the new cases.

The timing has contributed to confusion. The ruling often cited in coverage was issued on June 23, 2025, in Bartz, not in a May 2026 lawsuit. In May 2026, authors who did not join or accept the settlement filed separate cases. Chang-rae Lee is identified as a named plaintiff in one of them, Kwon et al. v. Anthropic. The source article said it could not find Wolff among the plaintiffs listed in five related cases as of August 2026.

Anthropic has disputed wrongdoing. In settlement materials, the company represented that neither the LibGen nor PiLiMi datasets, nor portions of them, were in the training corpus of any commercially released large language model. That statement sits alongside the court record’s findings about downloads and training-related use of portions of the collections in some models. The settlement also resolved claims about past conduct only within its stated scope; it did not grant a general license to use books in future training.

““spectacularly so””

— Judge William Alsup, in the June 2025 Bartz order

Open Questions in the New Cases

The new lawsuits remain unresolved allegations. A hearing on Anthropic’s motion to dismiss is scheduled for December 17, 2026, but the available source material does not establish whether the claims will proceed to trial or how the court will rule. No judge ruled in May 2026 that Anthropic illegally trained Claude on the books named in the new suits.

There is also a distinction between the court’s findings about the downloaded collections and Anthropic’s statement about commercially released models. The settlement representation says those datasets and their portions were not in the training corpus of any commercially released model; the cited 2025 record describes portions of the pirated collections being used to train some models. The precise scope and reconciliation of those statements are not established by the material provided. The reported book list and the status of individual authors’ claims also require care: the source says it did not verify the list against the settlement database.

December Hearing and Case Scope

The next scheduled court milestone is the December 17, 2026 hearing on Anthropic’s motion to dismiss the opt-out authors’ claims. The court’s handling of that motion will indicate which claims, if any, can continue, but it will not by itself establish liability. Settlement payments also have scheduled installments, including amounts due by September 27, 2027, according to the settlement terms summarized in the source.

Readers should distinguish future court rulings from statements by either side and from newspaper reports identifying affected books. Until the court rules, the new plaintiffs’ allegations remain claims, while the 2025 ruling and the approved settlement concern the earlier Bartz case and its defined scope.

Key Questions

Did a court rule that Anthropic illegally trained Claude on pirated books?

Not in the way that wording suggests. In June 2025, Judge William Alsup found that training Claude on books was fair use, but that downloading and keeping pirated book copies was not. New author lawsuits filed in 2026 have not been decided.

How much is the settlement?

The court approved a $1.5 billion settlement in July 2026, covering owners of roughly 500,000 registered books downloaded from LibGen or PiLiMi. The source describes the payment as about $3,000 per book before fees.

Does the settlement cover every author or every future use of books?

No. It covers defined claims over past conduct through August 25, 2025, for the books and claimants within the settlement. The source says claims about Claude’s outputs are not released and the settlement is not a general license for future training.

What does Anthropic say about the pirated collections?

Anthropic denies wrongdoing and represented in settlement materials that the LibGen and PiLiMi datasets, including portions of them, were not in the training corpus of any commercially released large language model. That is the company’s statement; the new lawsuits remain pending.

When could the new lawsuits advance?

A hearing on Anthropic’s motion to dismiss is scheduled for December 17, 2026. The court has not yet ruled on the new claims.

Source: ThorstenMeyerAI.com