Anthropic, the maker of the AI model Claude, said in a Thursday brief that the circuit court should hear the case now and resolve important questions about fair use while several other pending federal cases challenge the use of copyrighted books to train large language models.
U.S. District Judge William Alsup approved the class on July 17, saying he thought it would be "straightforward" for the authors' to prove that Anthropic pirated millions of books in an effort to train Claude quickly.
"Despite the complexities and stakes, the district court here rushed to certify a massive class threatening hundreds of billions of dollars in statutory damages, without any reliable means to identify class members or adjudicate their claims," Anthropic's brief said. "The certification decision also impermissibly purported to resolve critical merits issues. And it did so by glossing over the individualized issues that would pervade a fair trial."
Those issues include ownership, registration, validity of the copyrights at issue and damages, according to Anthropic. Judge Alsup said in his class certification order that it was possible an issue could arise as to a copyright's true ownership, such as an author and publisher each claiming they should recover part of an award.
But the judge said that, in his experience, this was "unlikely to occur except in the rarest of instances." Anthropic argued that the judge's citation of his own experience is not the rigorous analysis required by Rule 23 of the Federal Rules of Civil Procedure, which governs class certification.
Anthropic said the certification had improperly foreclosed its fair use defense altogether and that the "apparent use of a class-certification order to resolve critical merits issues contravenes Supreme Court precedent."
Judge Alsup had impermissibly waded into the merits of Anthropic's fair use defense while certifying the class, and the certification should be overturned so the fair use issue can be assessed again, Anthropic said.
Earlier in the case, Judge Alsup granted partial summary judgment to Anthropic, saying that its use of legally purchased books to train Claude was transformative and protected by fair use, but that its downloading of books from online pirate libraries would go to trial.
Anthropic said Judge Alsup's order raised a "host of individualized factual issues" that were then brushed away by the class certification when he said fair use would be treated "by common evidence and common methods" were it to come up at trial.
The company said the certification presents a "death-knell" situation which puts it under huge pressure to settle with the authors rather than defend itself from a potentially massive damages award. Anthropic said it could be on the hook for hundreds of billions of dollars. It currently operates at a loss and said posting an appeal bond could be impossible.
According to Anthropic, Judge Alsup had encouraged settlement talks by deviating from a typical bar on such talks before class certification and expedited the case's schedule.
Counsel for the authors did not immediately respond to requests for comment.
The authors are represented by Justin Nelson, Alejandra Salinas, Rohit Nath, Michael Gervais, Jordan Connors, Samir Doshi and Molly Karlin of Susman Godfrey LLP and Rachel Geman, Jallé Dafa and Reilly Stoler of Lieff Cabraser Heimann & Bernstein LLP.
Anthropic is represented by Kathleen Hartnett, Ephraim McDowell, Alexander Kasner, Elias Kim and Allison O'Neill of Cooley LLP, Daralyn Durie of Morrison & Foerster LLP and Douglas Winthrop of Arnold & Porter Kaye Scholer LLP.
The case is Andrea Bartz et al. v. Anthropic PBC, case number 25-4843, in the U.S. Court of Appeals for the Ninth Circuit.
--Editing by Dave Trumbore.
Correction: A previous version of this story included a mispelled name. The error has been corrected.
For a reprint of this article, please contact reprints@law360.com.
