San Francisco U.S. District Judge Araceli Martínez‑Olguín halted the final approval of Anthropic's proposed $1.5 billion settlement with a class of authors who claim the company trained its Claude AI models on millions of illegally downloaded books. The judge, who took over the case from Judge William Alsup earlier this year, said she needs a clearer picture of how the settlement fund will be divided before she can sign off.
The lawsuit, Bartz v. Anthropic, was brought by thriller writer Andrea Bartz, nonfiction author Charles Graeber and journalist Kirk Wallace Johnson. They allege Anthropic scraped more than seven million titles from shadow libraries such as Library Genesis and Pirate Library Mirror to feed its large‑language models. If the settlement proceeds, roughly 480,000 works would qualify for payment, with each work slated to receive about $3,000 after fees – the highest per‑work award ever recorded in a U.S. copyright case.
Class counsel has already trimmed its fee request from 15 % to 12.5 % of the total fund. The proposed order also lists about $3 million in expenses, an $18.22 million cost reserve and $50,000 service awards for each of the three lead plaintiffs. Judge Martínez‑Olguín said the filing lacked sufficient detail on those line items and asked the lawyers to break down the numbers further.
Objections to the deal have surfaced from multiple angles. Some authors argue the settlement amount is too low given the scale of the alleged infringement. Others contend the notice campaign nudged class members toward filing claims instead of opting out. Mexican novelist Laura Esquivel, famous for "Como agua para chocolate," said she received notice three weeks after the deadline and never saw a Spanish‑language notice. Indie writer Victoria Pinder complained that authors who registered many books under a single copyright group would receive only one share, rather than a share per title, a practice class counsel described as avoiding a "windfall."
The legal framework behind the settlement stems from a June ruling by Judge Alsup, which held that training AI on books acquired legitimately qualifies as fair use because it is "exceedingly transformative." By contrast, the court deemed the construction of a "central library" from pirated copies as infringing, a distinction that forced Anthropic to the negotiating table.
Anthropic, which is courting a $30 billion financing round at a $900 billion valuation, sees the settlement as a one‑time charge amid a broader push for private‑equity investment. The company has publicly disagreed with the underlying fair‑use ruling but chose to settle to move past the litigation. Until Judge Martínez‑Olguín issues a further order, the settlement remains the largest proposed copyright payout in American history, but its final shape is uncertain.
The pause underscores the growing scrutiny of AI firms' data‑training practices. As more publishers and authors bring lawsuits against tech companies, courts are being asked to balance the promise of artificial‑intelligence innovation with the rights of creators whose works fuel that progress.
Este artigo foi escrito com a assistência de IA.
News Factory SEO ajuda você a automatizar conteúdo de notícias para o seu site.