Judge Approves Anthropic's $1.5 Billion Settlement Over Pirated Books Used to Train Claude

A San Francisco federal judge signed off Monday on Anthropic's $1.5 billion settlement with a class of authors who accused the AI company of stealing their books to train its Claude chatbot. It's the largest known copyright settlement in U.S. history, according to the NY Post.

U.S. District Judge Araceli Martinez-Olguin granted final approval, rejecting arguments from some authors that the deal was too small, the Post reported. Now-retired Judge William Alsup had already given preliminary approval last September.

Authors sued Anthropic in 2024, claiming the company downloaded pirated versions of their books to teach Claude how to respond to prompts. Alsup ruled last June that training an AI on copyrighted books can qualify as fair use. But he also found Anthropic broke the law by hoarding more than 7 million pirated books in what the company called a "central library" that wasn't necessarily even used for training, per the Post and Arab News.

That distinction mattered. Fair use covered the training. It did not cover just grabbing pirated copies and stockpiling them. A trial to determine damages for the piracy was set for last December, with potential damages running into the hundreds of billions of dollars if it had gone to a jury. Anthropic settled instead.

What Authors Actually Get

Under the terms, rightsholders with registered works swept into the case are set to receive about $3,000 per eligible book, according to Stocktwits. Anthropic also has to destroy the original pirated files and duplicates it pulled from shadow libraries like Library Genesis and Pirate Library Mirror, per Stocktwits.

An attorney for the authors said in court that rightsholders filed claims covering more than 92% of the roughly 480,000 works included in the settlement, according to the Post and Arab News. Devdiscourse put the figure at 91% of authors involved. Either way, the overwhelming majority of eligible authors are in.

The judge reserved more than $101 million of the settlement pool for attorneys' fees, according to Devdiscourse. That's a significant chunk of the pot going to lawyers, not authors, and it's part of why some writers pushed back.

The Objections Are Fair to Raise

Some authors argued the settlement wasn't big enough, overpaid plaintiffs' attorneys, or wrongly excluded certain copyright owners, according to both the Post and Arab News. That's a legitimate complaint. Anthropic is a company valued in the tens of billions of dollars, built in part on a foundation of pirated books, according to Alsup's own finding. A one-time payout of roughly $3,000 per book, split against $101 million in legal fees, is a rounding error for a company backed by Amazon and Alphabet.

But the judge weighed those objections and approved the deal anyway. Justin Nelson, the authors' lead attorney, called it "the largest known copyright recovery in history" and said his team is "gratified by the Court's ruling" and looking forward to distributing payments, per the Post.

What This Doesn't Settle

The settlement only resolves Anthropic's past downloading and training conduct. It is not a license for future use. Authors retain the right to sue over future model training or new generated outputs, according to Stocktwits.

Some authors and publishers already opted out of the class entirely and are pursuing separate lawsuits against Anthropic, which remain ongoing, according to Arab News and Stocktwits. So this isn't the end of Anthropic's legal exposure on copyright. It's the end of one specific chapter.

This is also the first major settlement among dozens of similar copyright suits filed against AI companies by authors and news organizations over how their work got used to train large language models, according to the Post. Every other company facing these suits, and there are many, just got a real data point on what a settlement costs and how courts are willing to draw the line between fair-use training and unlawful piracy of source material.

The open question now is whether $1.5 billion becomes the going rate other AI companies use as a benchmark in their own pending cases, or whether courts start pushing for bigger numbers once more of these disputes reach trial instead of settling first.