Anthropic's $1.5B Book Piracy Settlement Wins Approval

TL;DR

US District Judge Araceli Martinez-Olguin granted final approval on July 20 to Anthropic's $1.5 billion settlement in Bartz v. Anthropic, the largest copyright recovery in US legal history
Roughly 500,000 books are covered, at about $3,000 per work, after Anthropic built a library of more than 7 million pirated titles pulled from Library Genesis and Pirate Library Mirror
Plaintiffs' attorneys were awarded $101 million of the $187.5 million they requested; over 91% of eligible authors and publishers filed claims
The deal resolves liability for illegal downloading only - Judge William Alsup's separate 2025 ruling that training AI on legally acquired books is fair use stands untouched, and still isn't binding on any other AI company

A federal judge in San Francisco signed off Monday on Anthropic's $1.5 billion settlement with a class of authors and publishers, closing the country's biggest copyright case against an AI company with the largest payout of its kind ever recorded. The ruling ends Bartz v. Anthropic, filed in August 2024 by authors Andrea Bartz, Charles Graeber and Kirk Wallace Johnson, but it settles a narrower question than the headline number suggests.

What the Judge Actually Approved

US District Judge Araceli Martinez-Olguin granted final approval to the deal, which covers close to 500,000 books at roughly $3,000 per work. Anthropic must also destroy the pirated files it acquired from shadow libraries Library Genesis and Pirate Library Mirror, along with any copies derived from them.

The case had been sitting with the court since Judge William Alsup gave preliminary approval last September, then bounced back after he called the initial version "nowhere close to complete" and pushed the parties to fix gaps in how affected authors would be identified and paid. Alsup has since retired; Martinez-Olguin inherited the file and rejected objections that the settlement was too small for the scale of the infringement.

"It is the largest known copyright recovery in history," said Justin Nelson, lead attorney for the plaintiffs' class. "We look forward to making distributions to the Class as promptly as possible."

Martinez-Olguin dismissed complaints about the settlement's size as "not grounded in a realistic assessment of the overall risks and rewards of a trial," a signal that the court saw the deal as a reasonable outcome given the underlying liability.

The Money

The court awarded plaintiffs' attorneys $101 million in fees, well under the $187.5 million they had asked for. Participation was high: more than 91% of eligible authors and publishers filed claims, according to the settlement administrator's reporting cited across multiple outlets covering the approval.

Detail	Figure
Total settlement	$1.5 billion
Works covered	~500,000
Payout per work	~$3,000
Attorney fees awarded	$101 million
Attorney fees requested	$187.5 million
Claims rate	Over 91%

Not everyone signed on. A group of authors and publishers opted out of the class and filed separate suits against Anthropic, arguing the per-work payout undervalues their work and that the claims process wrongly excluded some copyright holders. Those cases proceed independently of Monday's ruling.

The Piracy the Money Is Actually For

The settlement traces back to how Anthropic built its training corpus, not to the act of training itself. Court filings from the underlying case established that the company obtained books through two distinct channels: buying and scanning physical copies, which the court treated as lawful, and downloading millions of titles from pirate sites for what Alsup's ruling described as a permanent "central library" that existed independent of any specific training run.

That distinction is why Anthropic ended up owing $1.5 billion despite winning the more consequential legal argument. Alsup ruled in 2025 that training Claude on legally acquired, copyrighted books qualifies as fair use under US copyright law, a decision widely read as protective of the AI industry's core practice. The piracy claim was separate: Anthropic amassed roughly 7 million pirated books through Library Genesis and Pirate Library Mirror before it ever became clear which of them would be used for training, and holding that library was infringement regardless of downstream use.

A wooden judge's gavel resting on its sound block against a plain background The settlement closes Anthropic's exposure for illegal downloading, but leaves the broader fair use question for training data untouched for every other AI company facing similar suits. Source: commons.wikimedia.org

Anthropic's deputy general counsel, Aparna Sridhar, framed the resolution around that same split. "We reached this settlement in 2025, after the court's landmark ruling that training AI on books is fair use under copyright law," she said, a statement that credits the settlement with resolving the piracy exposure while treating the fair use win as the more durable outcome for the company.

The scale of what Anthropic actually pirated has been documented before this settlement. Researchers have shown Claude and other frontier models can reproduce copyrighted novels from memory almost word for word, evidence that shadow libraries like Library Genesis fed directly into training corpora across the industry, not just at Anthropic. A broader look at how AI labs sourced their training data found similar patterns of pirated and scraped material at nearly every major lab.

A library room with wooden bookshelves lined with fiction and non-fiction titles Roughly 500,000 books are covered by the settlement, each entitling its author or publisher to about $3,000. Source: commons.wikimedia.org

What It Does Not Tell You

The number that made headlines, $1.5 billion, obscures how little of the surrounding legal fight this settlement actually resolves. Alsup's fair use ruling was never appealed as part of this case, and it isn't binding precedent for any court hearing a similar claim against another company. Judges weighing near-identical lawsuits against Google, Meta, Midjourney and OpenAI remain free to reach different conclusions about whether training on copyrighted text is fair use at all, whether the works were legally acquired, or both.

The settlement also doesn't establish a market rate for AI training data. The $3,000-per-work figure reflects the specific facts of illegal downloading, not a licensing price that other authors could point to when negotiating with AI companies from now on. And it does nothing to resolve the parallel legal fights other publishers are pursuing against AI companies over scraping and reuse in different contexts, including CNN's active copyright suit against Perplexity over news content rather than books.

Congress hasn't settled the underlying question either. A 300-page Senate discussion draft from Senator Marsha Blackburn would remove the fair use defense for AI training completely, which would undercut the legal foundation Anthropic just relied on to limit its own liability. That bill remains a draft, but it's a reminder that Monday's ruling settles one company's past conduct, not the rules that will govern the next one.

Anthropic gets to close its books on a two-year piracy case for a fixed, known cost, which is itself valuable to a company preparing for further fundraising and scrutiny. But the $1.5 billion changes nothing about whether AI training on copyrighted text is legal when the underlying copies were obtained honestly, and that fight is still being argued, case by case, in front of judges who owe this ruling nothing.

Sources: