In a July IP Hot Topic, we wrote about a pivotal summary judgment ruling in Bartz v. Anthropic that added another data point in the newly forming fair use landscape for copyright actions against GenAI companies. In that case, the court held that Anthropic’s use of books purchased or scanned from print to train its LLM Claude did qualify as “transformative” and thus constituted fair use, but that Anthropic’s collection and retention of a centralized library of pirated books was not protected under fair use. This bifurcated ruling left liability as to the piracy‑based conduct intact and set the stage for a damages trial. However, this week, Anthropic and the authors jointly filed for a class‑wide settlement after mediation. The settlement filing indicates that the parties intend to finalize the settlement by early September.
Three compelling factors may have driven both parties to settle:
- Damages Exposure: The bifurcated ruling left Anthropic with significant exposure to statutory damages for the works it had allegedly pirated. In the likely event that Anthropic’s infringement were found to be willful, these damages could have been $150,000 per work. With Anthropic having pirated at least seven million books, damages had the potential to run close to a trillion dollars.
- Damaging Publicity: With Anthropic’s use of pirated works taking center stage, Anthropic may have been eager to avoid fact discovery and the uncomfortable disclosures and publicity that could have resulted.
- Damaging Precedent: The court’s summary judgment decision foretold the potential for precedent neither party wanted. For authors, this court had strengthened the fair use doctrine for AI training data, and the court’s effusive praise of the technology might have dimmed the plaintiffs’ prospects. For Anthropic and other GenAI companies with a stake in the outcome, the court’s hostility to the use of pirated works in training data may have discouraged further involvement in this litigation. With much of this area of law being defined in real-time, neither party may have wanted to see this case go up on appeal and have precedent set in stone that would damage their respective industries. For example, settlement minimizes the risk to Anthropic of a potential final judgment and affirmation on appeal of the Bartz summary judgment holding that the use of pirated copies, even if immediately used for the transformative use and thereafter discarded, would not qualify as fair use. Instead, Anthropic may seek to take a precedent-setting course in a court more lenient to that issue, such as the Kadrey et al v. Meta Platforms, Inc. court.
While this settlement delivers closure and compensation for the plaintiff authors and prevents existential ruin for Anthropic, it also preserves uncertainty in the evolving fair use jurisprudence for generative AI. Practitioners should monitor the upcoming preliminary-approval hearing (September 8), review the detailed settlement terms once filed, and assess how this compromise may influence counsel, clients, or class claims in analogous AI-copyright matters.
