Context: Among the dozens of AI-related copyright infringement lawsuits that are presently pending (most of them in the U.S.), Bartz v. Anthropic in the Northern District of California is one of the more interesting ones. It’s a class action on behalf of book authors against an AI provider that is known for its Claude AI system, which excels at producing long texts, and has received many billions of dollars in funding and cloud computing resources from Amazon and Google. Senior District Judge William H. Alsup is now grappling with the question of whether the use of pirated material can be cured by the fair use exception to copyright infringement (April 4, 2025 ai fray article, in which we also made reference to the book authors’ motion for class certification).
What’s new: On Thursday (April 17, 2025), Anthropic filed its opposition to the class certification motion. The defendant argues that a copyright class relating to 5 million books is not manageable and that the questions are too distinct to be resolved in a class action. As part of that argument, Anthropic warns against “[t]he prospect of ruinous statutory damages—$150,000 times 5 million books”: that would mean $750 billion.
Direct impact: It is true that intellectual property rights are harder to enforce through class actions than consumer or shareholder rights, but as we discuss below, Anthropic’s arguments against class certification are not very convincing.
Wider ramifications: Various other class actions by right holders against AI providers are pending in the same district and in others, particularly also in the Southern District of New York.
Here’s Anthropic’s opposition brief, followed by further commentary:
The $750,000,000,000 doomsday scenario
Even if we assumed just for the sake of the argument that the mind-boggling number could be the ultimate outcome, it would not, in and of itself, preclude class certification. For example, the largest class-action settlement in the U.S., the Tobacco Master Settlement Agreement (Wikipedia), had a floor of $206 billion. That one was struck in 1998, so adjusted by inflation it would now amount to approximately $400 billion. It affected dozens of defendants and payments were made over a period of 25 years, but it shows that the potential damages award is not a knock-out criterion.
Statutory damages are damages that a U.S. court can determine without having to conduct a costly and lengthy damages trial with a jury where experts calculate potential lost profits or reasonable royalties). For a copyright infringement, the maximum amount if $150,000, and that one will be reached only if willful infringement is proven. That is, of course, Anthropic’s problem: it may win or it may lose, but it can’t possibly argue that it accidentally trained its Claude AI system on pirated books.
That said, the court would obviously know that a $750B damages award would drive Anthropic into bankruptcy.
Anthropic’s insistence on individualized proof
Anthropic’s lawyers argue that any copying of a given book could be de minimis (below the threshold where it starts to matter), fair use or unlawful. Here, however, the authors argue that their book were illegally used for LLM training, and it is hard to see how Anthropic could show that any given book was de minimis.
The plaintiffs point to license deals for this type of large-scale use of copyrighted material that involve a fixed per-book amount. While some book authors may rightly argue that they theoretically deserve more than others, large license deals (which also applies to patents, particularly when they are licensed through pools) offer transactional efficiencies for both sides at the expense of highly individual deal terms.
The brief quotes the 2023 Ninth Circuit decision in Hanagami v. Epic, but that one actually held a small number of dance moves to be potentially copyrightable regardless of quantitative considerations. In that decision, the appeals court essentially said it’s not just about quantity. This case here, however, is about books, and any thresholds for copyrightability or infringement will be a non-issue. Even large quantities of text may be uncopyrightable, but that would apply to telephone directories or other data collections devoid of expressive elements.
Many book authors use AI tools, but that doesn’t mean they don’t want to get paid
One section of Anthropic’s brief has the following headline:
“Plaintiffs are not adequate class representatives because many rightsholders use LLMs and do not object to the use of their works in training.”
This is a non sequitur. It is unsurprising that surveys show many people, including many book authors, use AI systems. Certain types of authors, such as academics, may be even more likely to do so than others. But it does not follow from someone’s use of a certain system that they necessarily condone copyright infringement (should the outcome of the case be that there is one). And even if so, they could simply opt out.
Overall assessment
There are also other arguments, such as that the book authors shouldn’t be entitled to injunctions because their objective is to seek compensation, or that there are sometimes ownership issues to sort out such as whether third parties actually have the exclusive right to grant licenses (as opposed to the authors). The latter is something that could be addressed at the point of distributing the payout, if and when they get there.
All in all, Anthropic’s arguments against class certification are not convincing. The case should be resolved on the merits.
Counsel
The following lawyers are named in the header of Anthropic’s brief:
- Arnold & Porter Kaye Scholer, San Francisco: Douglas A. Winthrop (lead counsel), Joseph Farris, Jessica L. Gillotte and Estayvaine Bragg.
- Arnold & Porter Kaye Scholer, Los Angeles: Angel T. Nakamura, Oscar Ramallo and Allyson Myers.
- Stanford professor Mark Lemley of Lex Lumina, who has been working for Google (a key Anthropic partner) for decades.
- Latham & Watkins, San Francisco: Joseph R. Wetzel and Andrew M. Gass.
The class certification motion was filed by the following lawyers:
- Susman Godfrey’s Justin A. Nelson (lead counsel), Alejandra C. Salinas, Collin Fredricks, Rohit Nath, Jordan W. Connors and J. Craig Smyser.
- Lieff Cabraser Heimann & Bernstein’s Rachel Geman (co-lead counsel), Jacob S. Miller, Danna Z. Elmasry, Daniel M. Hutchinson and Reilly T. Stoler.
- Cowan DeBaets Abrahams & Sheppard’s Scott J. Sholder and CeCe M. Cole.