Anthropic has won an important round in a US copyright fight over artificial intelligence training, but the ruling does not end the case. Senior district judge William Alsup found that training AI tools on copyrighted works can qualify as fair use, while also allowing authors to pursue claims tied to pirated copies of books.
The result is a split decision with consequences beyond one company. It gives AI defendants a major fair use argument, but it also draws a hard line around how training materials are obtained and stored.
What the court decided about AI training
The case is a class action brought by book authors who said Anthropic violated copyright by using their works without permission. It was first filed in August 2024 in the US District Court for the Northern District of California.
In a summary judgment order released late Monday evening, Alsup wrote: “The training use was a fair use.” The court treated the training process as legally different from simply replacing the original books in the market.
One key issue was whether the use was “transformative.” In copyright law, that question asks whether the use creates something new rather than acting as a substitute for the original work. Alsup wrote: “The technology at issue was among the most transformative many of us will see in our lifetimes.”
Chris Mammen, a managing partner at Womble Bond Dickinson who focuses on intellectual property law, described the order as the first major generative AI copyright ruling to address fair use in detail. He said Alsup found that training an LLM is transformative use, even where there is significant memorization.
Why the win is limited
Anthropic is the first artificial intelligence company to win this kind of fair use battle, but the ruling leaves a major problem unresolved. Alsup held that the authors can still take Anthropic to trial over pirating their works.
The order says Anthropic later moved to training on purchased copies of books. But before that shift, it had collected and kept a large library of pirated materials.
“Anthropic downloaded over seven million pirated copies of books, paid nothing, and kept these pirated copies in its library even after deciding it would not use them to train its AI (at all or ever again). Authors argue Anthropic should have paid for these pirated library copies. This order agrees,” Alsup writes.
The order concludes: “We will have a trial on the pirated copies used to create Anthropic’s central library and the resulting damages.”
That distinction matters. The court accepted fair use for training, but it did not accept piracy as a legally excused step in preparing a training library. Alsup wrote: “The downloaded pirated copies used to build a central library were not justified by a fair use.” He added: “Every factor points against fair use.”
How the book library became central to the case
During discovery, Alsup learned that Anthropic had partly relied on downloading large collections from pirated databases like Books3 to gather books for training AI tools like Claude. The order names Anthropic cofounder Ben Mann and describes several acquisitions.
According to the order, Mann downloaded the entirety of Books3 back in winter 2021. In June 2021, Mann downloaded at least five million copies of books from Library Genesis, or LibGen, which he knew had been pirated. In July 2022, Anthropic downloaded at least two million copies of books from the Pirate Library Mirror, or PiLiMi, which Anthropic knew had been pirated.
Those facts are now the basis for the remaining trial track. The fair use decision may help Anthropic and other AI companies on the training question, but the alleged pirate library creates a separate damages risk.
The lowest statutory damage for this type of copyright infringement is $750 per book. Alsup noted that Anthropic’s pirate library consisted of at least 7 million books, creating potential court-imposed penalties in the billions. There is no trial date set yet.
Why other AI copyright cases will watch closely
The ruling arrives while dozens of other AI copyright lawsuits are moving through the US legal system. In almost all of them, AI company defendants are making fair use arguments. Alsup’s decision is therefore likely to shape how those arguments are framed, especially in cases where piracy is not a factor.
The legal context is still unsettled. Before this order, there had only been one summary judgment decision in an AI copyright case. In Thomson Reuters v. Ross, a judge found that the AI startup Ross’s training on Westlaw materials was not fair use, and that case is already headed to an appeals court.
Alsup also has notable experience with fair use questions because he presided over the initial trial in Google v. Oracle, a tech and copyright case that later went before the Supreme Court.
Fair use advocates are already treating the Anthropic decision as a victory. Adam Eisgrau, the senior director of AI, Creativity, and Copyright Policy at Chamber of Progress, said the ruling should guide other courts assessing generative AI training on copyrighted material.
Anthropic spokesperson Jennifer Martinez told WIRED the company was pleased with the ruling, calling it “consistent with copyright's purpose in enabling creativity and fostering scientific progress.” Lawyers for the plaintiffs declined to comment.
The practical takeaway
The decision separates two questions that are often blended together in public debates about AI copyright. One question is whether copyrighted works can be used to train an AI system under fair use. In this case, the court said yes.
The second question is whether a company can build or keep a library of pirated works and rely on fair use to excuse that conduct. In this case, the court said no.
That makes the ruling both a major win for Anthropic and a serious warning. AI training may receive fair use protection in some circumstances, but the source and handling of training materials can still create major legal exposure.