Judge Approves Anthropic’s $1.5 Billion Copyright Settlement Over Pirated Books

A federal judge in San Francisco has granted final approval to Anthropic’s $1.5 billion settlement with authors and publishers whose copyrighted books were downloaded from pirate libraries as the artificial intelligence company built the systems behind its Claude chatbot.

U.S. District Judge Araceli Martínez-Olguín approved the agreement Monday, resolving class claims involving 482,460 works and bringing the lawsuit against Anthropic to an end. The court described the agreement as the largest copyright class action settlement in history. Payments are expected to average about $3,000 for each covered work before fees and costs.

More than 91% of the works covered by the settlement had been claimed as of April 16, representing 440,490 books. Only copyright owners whose works appear on the settlement list are included, meaning the agreement does not cover every book Anthropic obtained while building its collection.

The dispute began in 2024 when authors Andrea Bartz, Charles Graeber and Kirk Wallace Johnson sued Anthropic, accusing the company of copying their books without permission. The case later centered on books Anthropic obtained from online pirate libraries while assembling a large digital collection used in part to train its artificial intelligence models.

Court records show Anthropic downloaded more than 7 million pirated books from sources that included Books3, Library Genesis and Pirate Library Mirror. The company kept the material in a central digital library, including copies it later decided would not be used to train its AI models.

That distinction became central to the case. In June 2025, U.S. District Judge William Alsup ruled that Anthropic’s use of copyrighted books to train its large language models qualified as fair use under copyright law. He also found that converting legally purchased print books into digital copies for a searchable internal library was protected.

Alsup reached a different conclusion on books downloaded from pirate sites and kept in Anthropic’s permanent library. The judge found that the company could not treat every pirated book as protected simply because some books were later selected for AI training. A trial had been set to determine damages tied to the pirated copies before the parties reached the settlement.

Copyright law gives authors and other copyright owners exclusive rights over certain uses of their work, including making copies. Fair use creates an exception that can permit copying without permission based on factors that include the purpose of the new use, the type of copyrighted work, how much was copied, and whether the new use harms the market for the original. Courts weigh those factors based on the facts of each dispute.

The Anthropic ruling treated AI training and the creation of a permanent digital library as separate uses. Alsup found the training process sufficiently different from the original purpose of the books to qualify as fair use on the record before him. Keeping pirated copies available for other purposes was different because Anthropic had obtained those books without paying for them and continued to store them even when they were no longer needed for training.

The ruling does not establish that every use of copyrighted material to train artificial intelligence is automatically protected. Fair use is decided case by case, and a different method of obtaining, copying, or using protected material can lead to a different result.

Copyright law can also allow copyright owners to seek statutory damages instead of proving the exact amount of money they lost. Awards generally range from $750 to $30,000 for each infringed work and can rise to as much as $150,000 per work when willful infringement is proven. With hundreds of thousands of works at issue, those per-work damages created substantial potential exposure if the piracy claims had proceeded to trial.

Several authors objected to the settlement, arguing among other things that the payments were too low or that the attorneys representing the class would receive too much. Martínez-Olguín rejected the challenges to the agreement, finding that continued litigation carried significant risks and could have delayed or eliminated any recovery.

The judge did reduce the requested attorneys’ fees. Lawyers for the class sought $187.5 million but were awarded about $101.6 million, representing roughly 6.8% of the settlement fund.

The settlement also does not resolve every possible copyright claim involving Anthropic. Class members release certain claims tied to past copying and other AI inputs, but the agreement does not release claims involving AI outputs or company conduct beginning on or after August 25, 2025.

Anthropic said after the ruling that the settlement followed the court’s earlier finding that training its AI models on books qualified as fair use and pointed to the high participation rate among covered authors and publishers. Some copyright owners opted out of the agreement and are pursuing separate lawsuits against the company.

Martínez-Olguín entered final judgment and dismissed the class action with prejudice, preventing the settled claims from being brought again. The court will retain authority over distribution of the settlement fund and any disputes involving implementation of the agreement.