A $1.5 billion settlement between Anthropic and a group of authors is at risk after U.S. District Judge William Alsup signaled he may reject the deal.
The case centers on allegations that nearly 465,000 books were illegally pirated to train Anthropic’s Claude chatbot.
Settlement under strain
At a San Francisco hearing, Alsup said he was reluctant to approve the agreement, remarking,
“We’ll see if I can hold my nose and approve it.”
He set a September 25 follow-up hearing, with deadlines on September 15 for a finalized book list and September 22 for a claims form.
The proposed settlement, announced just days earlier, would pay about $3,000 per pirated book. It was meant to avoid a December trial.
Pirated books at the center
Alsup’s June ruling found that using copyrighted books for AI training is not automatically illegal, but that Anthropic had pulled millions of titles from pirate sites. Authors’ attorney Justin Nelson confirmed the list now includes about 465,000 works. Alsup demanded guarantees that the number will not grow.
Concerns over fairness and influence
Alsup raised concerns about whether all eligible authors would be notified and whether trade groups like the Authors Guild and Association of American Publishers were pressuring writers to accept the deal. “I have an uneasy feeling about all the hangers on in the shadows,” he said.
Authors Guild CEO Mary Rasenberger and AAP CEO Maria Pallante both attended the hearing. Pallante later called the judge’s timetable “troubling” and warned his approach could spark new conflicts between authors and publishers instead of resolving them.
Authors push back
Lead plaintiffs Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson attended but did not speak in court. Before the hearing, Johnson called the settlement “the beginning of a fight on behalf of humans that don’t believe we have to sacrifice everything on the altar of AI.”
Nelson argued the high-profile case has already attracted broad attention, ensuring fairness in distributing funds. He stressed this is not “an under-the-radar warranty case.”
With Alsup openly skeptical, the $1.5B deal could collapse, sending Anthropic and the authors toward a December trial that would directly test whether large-scale AI training on pirated books constitutes copyright infringement.
