AI IP Year in Review – The Market Harm Dilemma: What Bartz and Kadrey Reveal About Fair Useʼs Future in Generative AI

In June 2025, two Northern District of California decisions—Bartz et al. v. Anthropic PBC and Kadrey et al. v. Meta Platforms, Inc.—revealed fault lines in the forming fair use terrain for generative artificial intelligence (GenAI) copyright infringement actions. While similar in facts and outcome, the Bartz and Kadrey courts reached their findings of fair use through different analyses.

In Bartz, Anthropic had placed pirated or purchased-and-destructively-scanned copies of books into a central research library and used some of them to train Anthropic’s large language model (LLM), Claude. Judge Alsup found the fair use doctrine covered training LLMs and converting purchased print library copies into digital library copies but not building a central library using downloaded pirated copies. The court’s fair use analysis relied more heavily on the purpose and character of using copyrighted works to train LLMs to generate new text and less on market harm.

A few days later, in Kadrey, Judge Chhabria gave greater consideration to the market effect of GenAI infringement. Meta had trained its LLM, Llama, on both publicly available data and potentially pirated information. The court found the use of plaintiffs’ books to train its LLM to be highly transformative and therefore fair. But the court emphasized that, in the fair use analysis, harm to the market for the copyrighted work is more important than the purpose for which the copies are made. While Kadrey, like Bartz, also granted summary judgment on fair use in favor of the GenAI defendant, the court noted it might have found otherwise but for these plaintiffs making “the wrong arguments and fail[ing] to develop a record in support of the right one,” particularly as it relates to market harm.

Update

In August, the parties in Bartz settled. The court’s holding that Anthropic’s collection and retention of a centralized library of pirated books did not qualify as fair use set the stage for a damages trial. But Anthropic and the authors jointly filed for a classwide settlement after mediation. Three compelling factors may have driven both parties to settle:

1. Damages Exposure: With Anthropic having pirated at least 7 million books, statutory damages for those works had the potential to run close to a trillion dollars.
2. Damaging Publicity: Anthropic may have been eager to avoid fact discovery and the uncomfortable disclosures and publicity that could have resulted.
3. Damaging Precedent: The court’s summary judgment decision foretold the potential for precedent neither party wanted.

Takeaways

• Transformative use is strongly favored when GenAI systems employ mitigating software to prevent infringing or overly presumptuous borrowing in their outputs.
• Market effect can be a necessary counterweight to transformative use to defeat a fair use defense.
• Market harm may be shown more effectively through market dilution than hypothetical loss of licensing fees, which may not be cognizable in certain courts.
• The use of pirated copies, even if immediately used for the transformative use and thereafter discarded, might not qualify as fair use.

Read our original IP Hot Topics on these issues:

A Clash in California: Judicial Tug of War in GenAI Fair-Use Cases

From Trillion-Dollar Risk to Resolution: Settlement in the Anthropic Authors/Class Action

2025 Summer Associate Wade Marshall contributed to this article.

This article appeared in the 2025 AI Intellectual Property: Analysis & Trends Year in Review report.