Don't Let the AI Training Fair Use Headlines Get You Down

It feels like every tech outlet raced to declare "AI has won!" following the twin rulings that, for different reasons, found training LLMs on copyrighted works to be transformative and ultimately fair use.

On June 23, in Bartz v Anthropic, Judge Alsup held that using lawfully acquired books to train Anthropic's Claude model is fair use under Section 107, and that converting purchased print copies to digital also qualifies as fair use. However, he made clear that downloading and retaining pirated books to build a permanent research library exceeds fair use, sending that issue to trial on damages. In other words, downstream fair use of AI training doesn't excuse the upstream copying and storage of pirated content.

Furthermore, his ruling on digitizing licensed books directly conflicts with long-standing 2nd Circuit precedent, which has established that wholesale digitization of books without permission falls outside fair use. This portion of the decision may not survive appeal.

Two days later, in Kadrey v Meta, Judge Chhabria agreed that feeding books into Meta's Llama model is "highly transformative." Still, he criticized the absence of any developed argument on market harm in Bartz's fourth-factor analysis. Chhabria explained that had the plaintiffs' attorneys presented evidence and reasoning on how training affects the market for their works, he might well have ruled in their favor.

He went on to write, "This is not a class action, so the ruling only affects the rights of these thirteen authors - not the countless others whose works Meta used to train its models. And as should now be clear, this ruling does not stand for the proposition that Meta's use of copyrighted materials to train its language models is lawful. It stands only for the proposition that these plaintiffs made the wrong arguments and failed to develop a record in support of the right one." This represents a significant critique of the plaintiffs' legal strategy.

He went on to say, "This case presents the question whether such conduct [AI training of copyrighted works] is illegal. Although the devil is in the details, in most cases the answer will likely be yes."

While the plaintiffs in both cases will almost certainly appeal, the broader takeaway is that Judge Chhabria has effectively laid out a roadmap for future plaintiffs to pursue infringement claims against AI companies that train on copyrighted works without authorization or compensation. At the same time, the positions taken by Judge Alsup are highly suspect and run afoul of well-established law.

Joe G. Naylor

President, Digital Media Licensing Association