# Anthropic's $1.5 Billion Book Piracy Settlement Gets Final Approval, Even as Training AI on Legally Bought Books Stays Fair Use > Source: > Published: 2026-07-23 23:08:47+00:00 # Anthropic's $1.5 Billion Book Piracy Settlement Gets Final Approval, Even as Training AI on Legally Bought Books Stays Fair Use ## Federal judge approves record settlement in AI copyright case A federal judge granted final approval Monday to Anthropic's £1.17 billion ($1.5 billion) settlement with authors over pirated books used to train its Claude AI chatbot, the largest copyright settlement in US history. The approval comes even as a separate 2025 ruling that training on legally acquired books is fair use remains untouched. US District Judge Araceli Martínez-Olguín signed the order in San Francisco, confirming the deal resolves claims tied to more than 482,000 pirated works, with roughly 91 per cent of eligible authors and publishers, over 440,000 books, having already filed claims. The settlement stems from Bartz v Anthropic, filed in 2024 after authors accused the company of downloading hundreds of thousands of pirated books from illegal libraries LibGen and PiLiMi to build its training library. Judge William Alsup, who oversaw the case before retiring, ruled in June 2025 that using books to train AI was fair use, but that acquiring them through piracy was not. That distinction remains legally intact and separate from Monday's payout. ## Why Buying Books Is Legal — But Stealing Them Isn't: The AI Copyright Distinction That Matters Alsup's 2025 ruling drew a hard line between purchased books that Anthropic digitised for storage, which counted as fair use, and pirated books downloaded from illegal libraries, which did not. He ruled that Anthropic's digitisation of purchased books for storage counted as fair use, but that downloading pirated books from illegal libraries did not. That fair-use finding on legitimately acquired training data was never part of this settlement and stands as separate precedent even after Monday's approval. Legal commentary around the case has framed the distinction as ending AI's Wild West era of unregulated training-data acquisition, shifting scrutiny from whether AI training itself is lawful to how the underlying data was obtained. ## What the Settlement Actually Covers Martínez-Olguín confirmed the deal pays roughly £2,340 ($3,000) per work, about four times the usual statutory minimum for copyright infringement, and covers authors including lead plaintiffs Andrea Bartz and Kirk Wallace Johnson. She reduced the attorneys' fee award from the requested £146.3 million ($187.5 million) to roughly £79.2 million ($101.6 million), calling a 12.5 per cent cut too high and settling on a multiplier of 3.75. The court overruled all 54 objections filed by class members and third parties, including requests for source attribution or deletion of Anthropic's models, and explicitly noted the settlement does not release Anthropic from future claims over what Claude actually generates. ## Reaction From Both Sides Plaintiff attorney Justin Nelson called the outcome the largest known copyright recovery in history, adding that his team looks forward to making distributions to the class as promptly as possible. Anthropic deputy general counsel Aparna Sridhar said the company was pleased that more than 91 per cent of authors and publishers covered by the settlement have claimed their share of the payment. Judge Martínez-Olguín herself wrote that the £1.17 billion deal provides substantial benefits to the class in light of the novel claims asserted, noting that success at trial was not assured, and a loss would have left the class with no recourse. ## The New Legal Battleground: Acquisition, Not Training Legal observers say the case reframes what future AI copyright litigation will actually turn on. The fight over whether AI training itself qualifies as fair use is increasingly settled, leaving acquisition, meaning where training data came from and whether it was paid for, as the live legal question heading into 2026. That question sits at the heart of pending suits including [The New York Times v OpenAI](https://www.ibtimes.co.uk/us-newspapers-sue-microsoft-openai-ai-training-1805836) and [Disney and Universal's case against Midjourney](https://www.ibtimes.co.uk/meta-copyright-lawsuit-ai-training-1802953), both of which hinge on how the underlying training material was sourced rather than on the act of training itself. With final judgement entered, the Bartz case is formally closed, though the court will continue monitoring how payments reach the roughly 440,000 authors and publishers who filed claims. Anthropic is also required to delete the pirated files it downloaded, while a small number of authors, roughly 350 who opted out, remain free to pursue independent claims against the company. © Copyright IBTimes 2025. All rights reserved.