The Anthropic Allegations Over Claude’s Training Books, In Context

🔍 Read the full analysis: The Anthropic Allegations Over Claude’s Training Books, In Context on ThorstenMeyerAI.com

New lawsuits filed by authors in May 2026 have renewed scrutiny of Anthropic’s use of books to develop Claude, but no court has ruled on those new claims. In a separate case, a judge in June 2025 found that training Claude on books was fair use while ruling against Anthropic’s downloading and retention of millions of pirated copies; a later $1.5 billion settlement resolved eligible claims over that past conduct.

The 2025 decision came in Bartz v. Anthropic, a case brought by three authors. Judge William Alsup distinguished between using books to train an AI model and obtaining copies through piracy. He found that Anthropic’s training on books was fair use under copyright law. He also found that scanning books Anthropic had purchased was fair use. But he ruled that keeping pirated copies in a permanent library was not. The order described more than seven million copies from collections including Library Genesis and the Pirate Library Mirror.

The ruling was a pretrial decision, not a jury verdict on every issue. On the pirated copies, the judge accepted the authors’ account of the facts; a jury had been expected to decide remaining questions, including damages. The case settled before that happened. The settlement, finally approved in July 2026, provides about $3,000 per registered book for roughly 500,000 books, with payments made in instalments. It also requires destruction of the downloaded collections and copies, subject to legal-hold obligations.

Anthropic denies wrongdoing. In settlement materials, the company represented that the LibGen and PiLiMi datasets, or portions of them, were not in the training corpus of any commercially released model. That statement is distinct from the court’s findings about downloading and keeping the books, and the settlement resolves specified past claims rather than granting a general licence for future training. The settlement also does not release claims concerning Claude’s outputs.

The Ruling Separates Training From Piracy

The distinction matters because the legal dispute is not a single question of whether “AI training” is lawful. The 2025 order treated the purpose of training and the way source material was acquired as separate issues: training on books could be fair use, while obtaining and retaining pirated copies could still create liability. That framing may shape arguments in other copyright cases involving AI companies, although it does not decide those cases.

The settlement also gives eligible authors a defined payment process for a large number of registered works, while leaving other disputes open. Authors whose claims were not covered, and writers concerned about model outputs or later training practices, cannot treat this settlement as a ruling on those separate issues. For readers, the central takeaway is that the court did not simply declare either all training lawful or all use of books unlawful.

Popular right now

Dyson V8 Cordless Vacuum

AmazonView Latest Price

Bose QuietComfort Wireless Bluetooth Headphones with Active Noise Cancellation, Mic, USB‑C Charging, Deep Bass, 24‑Hour Battery – White Smok

AmazonView Latest Price

LG C2 Series 65-Inch OLED evo 4K Smart TV (OLED65C2PUA, 2022)

AmazonView Latest Price

As an affiliate, we earn on qualifying purchases.

How the Cases Reached This Point

Bartz v. Anthropic was filed in 2024 by three authors. In its June 23, 2025 order, the court assessed Anthropic’s use of books in training and its handling of copies. The judge described training as creating something different from the source books, but treated the acquisition and permanent storage of pirated files differently. Anthropic’s settlement later ended the covered claims without a jury trial on damages.

In May 2026, authors filed separate lawsuits, including Kwon et al. v. Anthropic. A report by The Stanford Daily identified several Stanford-affiliated writers whose books it said were among the affected copies. The source material for this article says Chang-rae Lee appears as a named plaintiff in one new suit, but it could not verify Tobias Wolff’s name on the plaintiff lists for five related cases as of August 2026. The book list has not been independently checked against the settlement database, so those identifications remain attributed to the newspaper’s reporting.

Anthropic’s securities filing has also been discussed alongside the copyright cases. According to the source material, its S-1 describes model behavior “resembling blackmail,” references debate over possible AI sentience as a potential regulatory concern, and notes that others have raised existential-risk concerns. Those disclosures do not say that Claude is sentient or that it will blackmail people or end humanity; they describe risks and concerns rather than established outcomes.

„spectacularly so“

— Judge William Alsup, in the June 23, 2025 order

New Author Claims Await a Ruling

No judge has ruled on the claims in the May 2026 lawsuits. Anthropic has sought dismissal, and a hearing is scheduled for December 17, 2026. The claims, the company’s defenses and the court’s eventual assessment remain unsettled. The available source material does not establish how the new cases will be resolved or whether they will proceed to trial.

There are also limits to what can be inferred from the older case and settlement. The court’s ruling addressed specific conduct and facts in Bartz; it was not a final ruling on every book named in news reports or every model Anthropic has developed. Anthropic’s settlement representation about commercially released models, the court’s findings about pirated copies, and the scope of the new plaintiffs’ claims are separate points. The reported book lists and the status of some individual authors require verification against court and settlement records.

December Hearing Will Test the Claims

The next stated court milestone for the new litigation is the December 17, 2026 hearing on Anthropic’s motion to dismiss. The judge’s decision on that motion will indicate whether some or all claims can proceed, but it will not by itself resolve the underlying copyright disputes if the cases continue. Further filings may clarify which works and alleged conduct are at issue.

For the settled Bartz claims, the settlement terms set out continuing payment instalments, including a final scheduled payment in September 2027, and require destruction of the specified pirated collections subject to legal holds. Those obligations are separate from the outcome of the new lawsuits. Until the court rules, claims in the new cases should be treated as allegations, not findings of liability.

Key Questions

Did the court rule that Claude training on books was illegal?

No. In the June 2025 Bartz order, Judge Alsup found that training on books was fair use. He separately ruled that downloading and permanently retaining pirated copies was not.

What does the $1.5 billion settlement cover?

It resolves specified past claims for eligible registered books downloaded from LibGen or PiLiMi. The settlement does not grant a general licence for future training and does not release claims about Claude’s outputs.

Have the lawsuits filed in 2026 been decided?

No. The claims remain unresolved. A hearing on Anthropic’s motion to dismiss is scheduled for December 17, 2026.

Did Anthropic say pirated books trained commercially released Claude models?

Anthropic represented in the settlement materials that the LibGen and PiLiMi datasets, or portions of them, were not in the training corpus of any commercially released model. That representation is distinct from the court’s findings about downloading and retaining pirated copies.

Source: ThorstenMeyerAI.com