🔍 Read the full analysis: What’s Proven About Anthropic’s Claude Training—and What Isn’t? on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
A federal judge’s 2025 ruling separated Anthropic’s use of books to train Claude, which the judge found was fair use, from the company’s downloading of millions of pirated books to build a permanent library, which the judge found unlawful. Anthropic later agreed to a $1.5 billion settlement covering about 500,000 books, while new opt-out author lawsuits have not been decided. The exact relationship between the pirated collections and commercially released models remains disputed.
A 2025 federal court ruling found that Anthropic’s use of books to train Claude was fair use, but that downloading more than seven million pirated book copies for a permanent library was not. The distinction matters as authors pursue new claims and the company’s $1.5 billion settlement covers past conduct involving about 500,000 books, not every dispute over Claude’s training or outputs.
In Bartz v. Anthropic, U.S. District Judge William Alsup ruled on June 23, 2025, that training Claude on books was fair use under copyright law. He also found that Anthropic’s purchase of used print books, followed by cutting and scanning them, was fair use. The ruling drew a different conclusion about copies the company downloaded from online collections: it described 196,640 books from Books3, at least five million from Library Genesis and at least two million from the Pirate Library Mirror. The judge said downloading those copies to retain a permanent library was not fair use.
The decision was a pretrial ruling, not a verdict after a trial on damages. For the piracy issue, Alsup accepted the authors’ account of the relevant facts; a jury was expected to decide remaining matters, including damages, before the parties settled. Anthropic agreed to pay $1.5 billion to resolve covered claims for roughly 500,000 registered books. The settlement provides about $3,000 per book before fees and requires destruction of the downloaded LibGen and PiLiMi books and copies, subject to legal-hold obligations.
The settlement did not settle every later author claim. Authors who opted out filed new lawsuits, and no court ruling on those claims is identified in the source material. A hearing on Anthropic’s motion to dismiss is scheduled for December 17, 2026. Separately, Anthropic told the settlement court that the LibGen and PiLiMi datasets, or portions of them, were not in the training corpus of any commercially released model. That statement is the company’s position; the settlement does not itself establish that all training-related questions are resolved.
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
What the Ruling Means for Claude
The distinction between training and acquisition is central to the dispute. The court found the use of books to train a model lawful in the circumstances before it, while finding that obtaining pirated copies to create a lasting collection was not. The ruling is not a blanket permission to use any copyrighted material, nor does it decide every possible claim about how a model was trained or what it produces.
The settlement puts a substantial price on the covered book-copying claims and provides a process for eligible rights holders to seek payment. It also leaves a live question for authors who opted out: what legal claims remain available and whether they can proceed. For publishers, writers and AI developers, the cases may shape how courts treat the sourcing and retention of training materials, a separate issue from whether model training itself qualifies as fair use.
Readers should also distinguish court findings from broader claims about Claude’s behavior. Anthropic’s securities filing describes certain model behavior as “resembling blackmail” and refers to public debate about AI sentience and existential risks. Those disclosures do not establish that Claude is sentient or that it blackmailed a real person. The reported blackmail example was a staged safety test, according to the source material.
Products Worth Considering
LG UltraGear 32GX850A-B 32-Inch 4K UHD Glossy OLED Gaming Monitor
Apple iPad Air (2025) 13‑inch Wi‑Fi 256 GB – Space Gray (M3 chip)
Apple iPad Air (4th Generation, 10.9-inch, Wi-Fi, 256GB) – Space Gray
As an affiliate, we earn on qualifying purchases.
How the Copyright Cases Developed
The case that produced the 2025 ruling was Bartz v. Anthropic, filed by three authors in August 2024. Judge Alsup’s June 23, 2025 order addressed several ways Anthropic obtained books: buying print copies and scanning them, downloading from online collections, and using books to train Claude. It treated those actions differently rather than deciding that all book use was either lawful or unlawful.
Anthropic later settled the covered claims for $1.5 billion. The settlement applies to past conduct through August 25, 2025, and expressly does not release claims concerning Claude’s outputs. It is not a licence for future training on any material. The company denies wrongdoing. According to its representation to the settlement court, the pirated LibGen and PiLiMi material was not in the training corpus of any commercially released large language model.
In 2026, authors who did not join the settlement brought new suits. The source identifies Chang-rae Lee as a named plaintiff in one, Kwon et al. v. Anthropic. It also reports that The Stanford Daily named several Stanford-affiliated writers whose books it said were among the pirated copies, including Tobias Wolff, Abraham Verghese, Daniel Mason, Hayagreeva Rao and Myra Strober. Those identifications are reported by the student newspaper; the source author says they did not independently check the book list against the settlement database. The source also says Wolff’s name was not found among plaintiffs listed in five related cases as of August 2026.
““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, in a representation to the settlement court
Which Training Claims Remain Open
The record supports a finding that Anthropic downloaded large numbers of pirated books and a separate finding that training on books was fair use in the case before Judge Alsup. It does not establish, by itself, precisely which pirated books were used to train which model versions, or settle every question about the data behind Claude products. Anthropic’s statement that the LibGen and PiLiMi material was absent from commercially released models should be presented as a company representation, not as an independently adjudicated fact.
The new opt-out lawsuits are unresolved. No May 2026 ruling found Anthropic liable in those cases, despite coverage that may conflate them with the June 2025 Bartz decision. The source material does not provide a court determination of which specific Stanford authors’ books appear in the settlement database, nor does it establish the final status of every author’s participation in related actions.
There is also a scope limit to the settlement: it releases covered claims about past conduct through August 25, 2025, while expressly excluding claims about model outputs. How courts will treat the new claims, and whether any evidence will establish a connection between particular copies and particular commercial models, remains unsettled.
The next scheduled court milestone identified in the source is the December 17, 2026 hearing on Anthropic’s motion to dismiss the opt-out authors’ claims. The judge’s decision on that motion may determine which claims proceed, but it will not necessarily resolve the underlying factual disputes.
Settlement administration and payments also remain relevant. The reported schedule includes instalments through September 27, 2027, while eligible rights holders seek payment for registered books. Further court filings, any ruling on the dismissal motion and developments in the settlement process could clarify the remaining disputes. Until then, the 2025 ruling, the settlement, Anthropic’s statements and the new authors’ allegations should be treated as distinct parts of the record.
Key Questions
Did a judge rule that Claude’s training on books was illegal?
No. In June 2025, Judge William Alsup found that training Claude on books was fair use in the case before him. He separately found that downloading pirated books to build a permanent library was not fair use.
What does Anthropic’s $1.5 billion settlement cover?
The settlement covers specified past claims involving about 500,000 registered books downloaded from LibGen or PiLiMi. It applies to conduct through August 25, 2025, and does not release claims about Claude’s outputs.
Did the settlement court confirm pirated books were used in commercial Claude models?
The source describes no such court finding. Anthropic told the settlement court that the LibGen and PiLiMi datasets, or portions of them, were not in the training corpus of its commercially released models. That is Anthropic’s representation, not a ruling resolving every data-use question.
Have the new lawsuits by authors who opted out been decided?
No ruling on their claims is identified. A hearing on Anthropic’s motion to dismiss is scheduled for December 17, 2026.
Does Anthropic’s filing say Claude is sentient or will blackmail people?
No. The filing, as described in the source, refers to model behavior “resembling blackmail”, a debate about AI sentience and concerns raised by others about existential risks. Those disclosures are not proof that Claude is sentient or has blackmailed a real person.
Source: ThorstenMeyerAI.com
