A U.S. federal judge has given final approval to Anthropic’s $1.5 billion settlement with authors who accused the AI company of violating their copyrights, Reuters reported. The ruling matters less because it rewrites AI law, and more because it draws a line AI companies have been trying to blur: training on books was treated as fair use, while keeping a centralized stash of pirated books was not.
According to Reuters, the court found that using published books to train AI models falls under fair use in this case. Anthropic’s problem was its library of 7 million pirated books, which the court said infringed the rights of authors and publishers.
The settlement is reportedly the largest payout ever in a copyright case. It is also the first settlement among a broader set of copyright fights aimed at AI companies over the material used to build and train their models.
The dollar figure looks severe until it is divided across the works at issue. Reuters reported that the payment amounts to a little more than $200 per title because of the number of books and claimants covered by the class action.
Fair use survived, the pirated library did not
Anthropic deputy general counsel Aparna Sridhar said the company reached the settlement in 2025 after the court ruled that training AI on books is fair use under copyright law. She said that ruling remains controlling and that more than 91% of covered authors and publishers have claimed their share of the settlement payment.
That is the uncomfortable split in the case for both sides. Authors and publishers get a record settlement over Anthropic’s pirated collection. AI companies also get a ruling they can point to when arguing that model training on published works does not, by itself, violate copyright law.
Some groups have declined to join the settlement and instead filed separate complaints against Anthropic, according to the report. Their decision suggests the deal has not ended the fight over how AI companies obtained training material, or how much copyright owners should be paid when those companies used unlawful copies.
Other AI copyright cases are still moving
The Anthropic settlement lands while other AI copyright disputes continue. Nvidia has been accused of using scripts in its NeMo framework designed to download books illegally. In that case, the company argued that its framework had non-infringing uses, but the court focused on specific tools inside the system rather than the framework as a whole.
Meta has also faced claims over training data. Court records reportedly said Meta staff torrented nearly 82TB of pirated books for AI training. Meta has argued that use of pirated material was legal if it did not seed the files.
Those cases are not resolved by Anthropic’s settlement. The decision does, however, give both sides a cleaner map of the battlefield. Courts may accept AI training as fair use while still treating pirated data acquisition and storage as ordinary copyright infringement. That is bad news for any company hoping the word “AI” turns a pirate library into research infrastructure.
This story draws on original reporting from Tom's Hardware.
