A federal judge in California has declined to rubber-stamp what was shaping up to be the largest copyright settlement in United States history. US District Judge Araceli Martinez-Olguin paused the approval process for Anthropic's $1.5 billion settlement with a class of authors and publishers, citing significant concerns raised by a subset of the very people it was supposed to help.
What Happened in Court
The settlement arose from litigation over Anthropic's use of copyrighted books to train its Claude family of AI models -- a practice that has drawn legal fire from authors, publishers, and rights holders across the technology industry. On Thursday, Judge Martinez-Olguin said she needed more time to assess why some class members were opposing or opting out of the deal before she could move forward with final approval.
Among the loudest voices of dissent: the legal fee structure. Under the proposed terms, the plaintiffs' attorneys stand to collect more than $320 million -- while the average participating author would receive approximately $3,000. Critics called the math deeply inequitable. "Every dollar that Counsel takes from the Settlement fund is one that is not given to those actually harmed," wrote Pierce Story, an author and objector whose works were among those swept into the training dataset.
Objectors Want More for Authors
Multiple objectors submitted letters to the court accusing the legal team of moving too quickly to finalize a settlement that prioritized attorney compensation over adequate relief for writers. Some class members claimed they were actively discouraged from raising concerns -- an allegation that, if substantiated, could threaten the settlement's legal standing under fair representation principles.
The judge's request that attorneys directly address these objections is unusual in class action proceedings of this scale, signaling that the court is not treating approval as a formality. The questions raised -- about fee proportionality, class representation, and adequacy of compensation -- are likely to significantly reshape the final terms.
Why Authors Are Pushing Back
The frustration among authors runs deep. AI training datasets have consumed millions of books, articles, and creative works without payment or consent, and many creators feel that a $3,000 lump-sum payment is a poor return for what they view as an ongoing commercial exploitation of their intellectual property. Some authors have argued the settlement should establish ongoing royalty structures rather than a one-time payment, especially given the continued commercial success of models like Claude.
Others have pointed to the speed of the process itself as evidence that class interests were subordinated to a desire for quick resolution. In the world of AI copyright litigation -- where parallel cases involving OpenAI, Meta, and Google are proceeding simultaneously -- there is significant pressure to reach landmark decisions before norms harden.
Why It Matters
This case is the most visible legal reckoning with AI training data practices to date. The outcome will shape what fair compensation looks like for human creators whose work underpins generative AI -- and whether settlement structures in future AI copyright cases will more meaningfully center the rights holders rather than the legal teams who represent them. The delay is a signal: courts are not ready to simply wave through AI industry settlements that do not hold up to close scrutiny.
Source: Ars Technica | Published: May 16, 2026