I’ve been closely following cases related to generative AI copyright as they unfold, which has been both interesting and unsettling, especially for those who followed Judge William Alsup’s ruling in Bartz et al. v. Anthropic PBC, which was delivered in June 2025.
In Bartz et al v. Anthropic PBC, the plaintiff authors alleged that their copyrighted books were unlawfully copied and stored in Anthropic’s “shadow libraries” and used to train its Claude AI models. Judge Alsup ruled that training an AI model on copyrighted works could qualify as fair use, but hoarding pirated copies in a centralized repository does not qualify as fair use. The case was adjourned for trial in December 2025, with billions or trillions of dollars at stake in potential liability.
Now, parties have filed a Notice of Settlement (the terms of settlement remain confidential at the moment), indicating their intent to execute a full settlement agreement by September 3, 2025. They have stipulated to stay discovery and vacate all existing deadlines and hearings.
Bartz et al v. Anthropic PBC was shaping up to be the first real test of fair use in the age of generative AI. By settling, Anthropic has avoided a catastrophic risk, but the rest of us are left without the clarity that a definitive appellate judgment might have offered.
I can’t help but see this as a relief on one hand, because companies and creators now have one less existential case looming overhead. However, it’s also a missed opportunity, as we still lack the robust judicial guidance that could anchor this debate for the decade ahead.
Further to this, a “Motion for Preliminary Approval” filed on September 5, 2025, discloses that Anthropic PBC has agreed to establish a settlement fund of $1.5 billion to resolve the claims brought by the plaintiff authors.
Notably, the structure of the fund is designed to account for the scale of the alleged infringement, with an initial allocation reflecting compensation for approximately 500,000 works. The settlement agreement also includes a mechanism for the fund to increase should additional rights holders be identified and come forward.
This proposed settlement, pending Judge Alsup’s approval, marks a pivotal moment. It underscores the substantial financial and legal implications for AI developers training models on copyrighted materials.
