The Anthropic book lawsuit has ended with a federal judge approving a $1.5 billion settlement for authors, publishers and other copyright owners. The case accused artificial intelligence company Anthropic of downloading hundreds of thousands of copyrighted books from online piracy sources while developing its Claude AI models.
U.S. District Judge Araceli Martínez-Olguín granted final approval to the agreement on July 20, 2026. The court described the settlement as fair, reasonable and adequate, considering the uncertainty the copyright owners would have faced if the case had proceeded to trial. It has been described as the largest known copyright class-action recovery in history.
How the Anthropic Book Lawsuit Began
The case, Bartz v. Anthropic PBC, was filed in the U.S. District Court for the Northern District of California in 2024. Authors Andrea Bartz, Charles Graeber and Kirk Wallace Johnson were among the original plaintiffs.
They alleged that Anthropic obtained copyrighted books from Library Genesis, commonly called LibGen, and Pirate Library Mirror, or PiLiMi. These websites have been accused of distributing books without permission from their authors or publishers.
The plaintiffs claimed Anthropic downloaded millions of files to create a permanent digital research library. Some of those books were allegedly used to train the large language models behind Claude.
Anthropic denied violating copyright law. The company argued that using books to train an AI system was transformative and therefore protected under the fair-use provisions of U.S. copyright law.
Important Fair-Use Ruling
Before the settlement was reached, U.S. District Judge William Alsup issued an important ruling concerning how Anthropic acquired and used the books.
The judge concluded that using books to train Anthropic’s AI models could qualify as fair use because the process was considered transformative. The models were not simply created to provide replacement copies of the original books.
The court also found that Anthropic could legally purchase printed books, remove their bindings and scan them into digital files for internal storage. Since the physical copies had been lawfully purchased, converting them into another format was treated as a permissible replacement of those copies.
However, the court reached a different conclusion about books downloaded from piracy websites. Judge Alsup ruled that creating a permanent library from unlawfully obtained copies was not protected merely because some of the books might later be used for AI training.
This distinction became central to the case: the purpose of AI training could be fair use, while obtaining the training material through piracy could still create copyright liability.
Details of the $1.5 Billion Settlement
The settlement covers approximately 482,000 eligible copyrighted works connected with the LibGen and PiLiMi files obtained by Anthropic. The official settlement process was open to legal and beneficial copyright owners, including qualifying authors and publishers.
The fund provides roughly $3,000 for each covered work before applicable legal fees, administrative expenses and ownership divisions. The amount for a particular book may be divided between its author, publisher or other rightsholders according to their contracts and legal interests.
By the time of final approval, claims had reportedly been submitted for approximately 91% of the eligible works. The deadline for submitting claims was March 30, 2026, while the deadline for requesting exclusion from the settlement was February 9, 2026.
The court approved more than $101 million in legal fees, representing about 7% of the total settlement fund. The settlement also includes payments for litigation expenses and service awards for the authors who represented the class.
Anthropic Must Delete Downloaded Files
The agreement requires Anthropic to destroy the original files downloaded from the identified LibGen and PiLiMi datasets within the period established by the final judgment. The company must also provide the required certification concerning the deleted materials.
The settlement resolves claims involving Anthropic’s past downloading, retention, scanning and use of the books included on the approved works list. It does not provide Anthropic with a licence to use those books in the future.
It also does not resolve possible claims based on material generated by an AI model. For example, a separate claim alleging that Claude reproduced protected text from a book would have to be considered on its own facts.
What the Decision Means for AI Companies
The Anthropic settlement does not establish that every use of copyrighted material for AI training is legal. The fair-use ruling applies to the particular facts presented in this federal district court case and does not automatically bind other courts.
However, the decision provides an important warning to AI developers. Even where the training process may be transformative, companies can still face substantial liability if they acquire copyrighted materials from unauthorised sources instead of purchasing or licensing them legally.
For authors and publishers, the $1.5 billion agreement provides compensation without the uncertainty of a lengthy trial and appeal. For Anthropic, it resolves major claims concerning its historical book library while leaving the earlier fair-use ruling largely intact.