🔍 Read the full analysis: What’s At Stake In The Allegations Over Claude’s Training Data? on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
TL;DR
A federal judge ruled in June 2025 that Anthropic’s use of books to train Claude was fair use, while finding that downloading millions of pirated book copies for a permanent library was not. Anthropic later agreed to a $1.5 billion settlement covering eligible books, and new opt-out authors’ lawsuits are still pending. The available source material does not establish that a court has ruled on those new claims or that Claude’s commercially released models trained on the pirated collections.
Authors who opted out of Anthropic’s $1.5 billion settlement over pirated book copies have filed new lawsuits, but no court has ruled on those claims, according to the source material. The dispute follows a June 2025 ruling that found training Claude on books was fair use while separately finding that Anthropic’s downloading of millions of pirated copies to build a permanent library was not.
The distinction matters because descriptions of the case as a ruling that Anthropic’s AI training was illegal collapse two different issues. In Bartz v. Anthropic, Judge William Alsup’s June 23, 2025 order said training on books was fair use. It also said scanning books Anthropic had bought was fair use. The judge found that downloading pirated books and retaining them in a library was not. That order was a pretrial summary-judgment decision; the pirated-copy issue proceeded toward trial, which did not take place after the settlement.
The order described more than seven million book copies from the Books3, Library Genesis and Pirate Library Mirror collections. The settlement, finally approved in July 2026, provides $1.5 billion for owners of roughly 500,000 registered books downloaded from LibGen or PiLiMi—about $3,000 per book before fees, according to the source. It also requires Anthropic to destroy the downloaded collections and copies, subject to legal-hold obligations. The agreement releases claims over past conduct through August 25, 2025; it does not release claims about Claude’s outputs.
Anthropic denies wrongdoing. In the settlement, the company represented that LibGen and PiLiMi materials were not in the training corpus of any of its commercially released large language models. The source also says the court record described portions of pirated collections being used to train some Anthropic models. These statements concern different points—the use of material in training and whether it was included in released models—and should not be treated as a court finding that a commercial Claude model was trained on those books.
From download to IPO
Five years, one ruling, no trial
The piracy happened in 2021 and 2022. The only ruling came in 2025. The new lawsuits are years away from a verdict.
- Jun 2021 · Jul 2022DownloadsAt least 5 million copies from LibGen (2021), at least 2 million from PiLiMi (2022); 196,640 from Books3 earlier.
- Aug 2024Bartz v. Anthropic filedThree authors sue Anthropic, on behalf of a class of book owners.
- 23 Jun 2025Judge Alsup's orderTraining on books: fair use. Scanning bought books: fair use. Pirated library copies: not fair use, to trial.
- Sep 2025$1.5 billion settlementNo trial. Anthropic denies wrongdoing; release covers conduct up to 25 Aug 2025 only.
- 13 + 15 May 2026Opt-out suitsCruz (28 plaintiffs) and Kwon (incl. Chang-rae Lee) join Cambronne and two publisher suits.
- Jul 2026Final approvalSettlement approved. Anthropic: no opt-out deal "above the Bartz claim value".
- 24 Sep 2026Motion to dismissAnthropic: claims too late, no infringing output shown, many works already released.
- 26–27 Oct 2026NowS-1 and the headlineIPO filing cites Bartz as a risk; The Stanford Daily names Stanford authors.
- 17 Dec 2026ScheduledHearingCourt hears the motion to dismiss.
- May 2027 · Jan 2028ScheduledDiscovery, then trial settingFact discovery closes 13 May 2027; trial-setting conference 11 Jan 2028.
Sources: Bartz v. Anthropic, order of 23 June 2025; settlement FAQ; joint case statement, 27 Aug 2026; Anthropic letter, 27 Jul 2026; case management order, 14 Sep 2026; Anthropic S-1, 26 Oct 2026; The Stanford Daily, 27 Oct 2026. Scheduled dates can move.
Powered by Thorsten Meyer AI
The case puts a practical boundary around two questions that are often blurred in debates over generative AI: how a company acquires books and whether using them to train a model is fair use. The judge treated the conduct differently. That means the 2025 order is not a blanket ruling that all AI training on copyrighted books is unlawful—or that any way of gathering training material is permissible.
For authors, the settlement offers compensation to people whose registered books qualify, but it does not resolve every copyright dispute. Authors who opted out are pursuing separate claims, and the settlement’s release is limited by its terms. For AI companies, the dispute carries financial and operational consequences: the settlement is substantial, requires destruction of specified downloaded files, and leaves open litigation over claims not covered by the agreement.
The outcome could also shape how publishers, creators and AI developers approach licensing and data sourcing. But the pending cases have not yet produced a ruling on the opt-out authors’ claims. Until then, the settled case provides a record of one judge’s treatment of specific conduct and facts, not a final answer to every question about copyright and model training.
Products Worth Considering
Logitech MX Keys S Wireless Keyboard, Graphite
Amazon Kindle Scribe Colorsoft 64GB (1st Generation)
LG 34GX90SA-W 34-inch Ultragear WQHD OLED Curved Gaming Monitor
As an affiliate, we earn on qualifying purchases.
How the Bartz Case Split the Issues
Bartz v. Anthropic was filed by three authors in August 2024. In the June 2025 order, Alsup described Anthropic’s training as transforming the books into something different and found that use fair. The order separately addressed Anthropic’s acquisition of books: the company had bought millions of used print books, cut them from their bindings and scanned them, which the judge also treated as fair use. By contrast, the judge found that keeping pirated copies downloaded from online collections was not fair use.
The decision did not determine all damages at trial. On the pirated copies, the judge accepted the authors’ version of the facts at the pretrial stage, and a jury was expected to decide remaining issues. The parties settled instead. The settlement now covers eligible book owners and specified past conduct, while expressly leaving claims about Claude’s outputs outside its release.
A separate Stanford Daily report named Stanford-affiliated authors it said had books among the pirated copies, including Tobias Wolff, Abraham Verghese and Chang-rae Lee. The source article cautions that it did not independently verify that reported book list against the settlement database. It also says Chang-rae Lee appeared as a plaintiff in one new suit, while the author could not find Wolff listed among plaintiffs in five related cases as of August 2026. Those details should be treated as reported, not as findings in the court order.
“neither the LibGen or PiLiMi datasets, nor any portions of those datasets, were in the training corpus of any of its commercially released large language models.”
— Anthropic, as represented in the settlement
What the New Claims Have Yet to Establish
The opt-out lawsuits remain unresolved. The source says new suits were filed in May 2026 and that no judge ruled on them that month; it identifies a December 17, 2026 hearing on Anthropic’s motion to dismiss. The allegations in those cases have not been established by a ruling.
It is also not clear from the supplied material exactly which books or authors are included in each new action, or how the court will address the claims. The reported list of Stanford authors has not been independently checked against the settlement database in the source account. Separately, the tension between the court record’s account of training and Anthropic’s settlement representation about commercially released models requires careful wording: the material does not establish that a particular released Claude model trained on the pirated collections.
The December Hearing and Settlement Payments
The next identified court milestone is the December 17, 2026 hearing on Anthropic’s motion to dismiss the opt-out authors’ claims. The court could then decide whether some or all of those claims proceed; the supplied material does not state what ruling is expected or when one might come.
The settlement is being paid in four instalments, according to the source: $300 million in October 2025, $300 million after final approval, $450 million by September 25, 2026, and $450 million by September 27, 2027. The final instalment remains scheduled for 2027. The new cases are separate from the settlement, and their outcome may clarify which claims remain available to authors who did not participate in it.
Key Questions
Did a judge rule that Claude’s training on books was illegal?
No. In June 2025, Judge William Alsup found that training on books was fair use. He separately found that downloading pirated copies to keep in a permanent library was not fair use.
What does the $1.5 billion settlement cover?
It covers eligible owners of roughly 500,000 registered books that Anthropic downloaded from LibGen or PiLiMi, and releases specified claims over past conduct through August 25, 2025. Claims about Claude’s outputs are not released by the agreement.
Have the new opt-out author lawsuits been decided?
No ruling is reported in the supplied material. It says a hearing on Anthropic’s motion to dismiss is scheduled for December 17, 2026.
Did Anthropic admit that released Claude models trained on pirated books?
The source says Anthropic represented in the settlement that LibGen and PiLiMi materials were not in the training corpus of its commercially released models. The supplied material does not establish that a particular released Claude model trained on those books.
Source: ThorstenMeyerAI.com
