Court Approves Anthropic's Record $1.5 Billion Settlement With Book Authors

A federal judge in San Francisco approved the final version of a settlement between Anthropic and book authors and publishers on Monday, closing out a dispute dating back to 2024 over the company training its Claude chatbot on pirated copies of their work. The settlement's value, $1.5 billion, makes it the largest known settlement in the history of US copyright law.

Where the Dispute Began

The Bartz v. Anthropic case reached court in 2024. Bestselling thriller author Andrea Bartz and two other writers accused the company, which is backed financially by Amazon and Alphabet, of illegally obtaining copies of their books from piracy sites to train the language model behind Claude. The lawsuit later grew into a class action covering hundreds of thousands of titles.

Judge William Alsup, who presided over the case until his retirement, drew a key legal line in June 2025. He ruled that training the model on legally purchased and scanned books qualified as transformative fair use. At the same time, he found that Anthropic had infringed the authors' rights by building an internal library of more than 7 million pirated copies downloaded from Library Genesis and Pirate Library Mirror, regardless of whether all of them were actually used to train the models.

The Settlement and Rejected Objections

Alsup granted preliminary approval of the settlement in September 2025, calling it "fair" while cautioning that the payout process itself would be logistically complex. At the final approval hearing in May 2026, several class members raised objections: that the list of eligible works undercounts the actual number of works by treating a single copyright registration as one work even though some registrations cover several separately published novels, that works published under a pseudonym were excluded from the settlement, which primarily hurts independent authors, and that the one-time payout is too low given Anthropic's ongoing profits from models trained on these books.

After additional briefing from both sides, Judge Martinez-Olguin rejected these objections and signed the final approval of the settlement on July 20, 2026. Payments are to go to the authors and publishers who filed claims against the list of roughly 500,000 covered works, at a pace the parties describe as as fast as possible.

This is the largest known amount ever recovered in the history of copyright disputes. We're looking forward to distributions to the class starting as soon as possible. - Justin Nelson, counsel for the authors

What It Means for the AI Industry

Because Anthropic chose to settle rather than appeal, Alsup's ruling separating lawful training from unlawful data acquisition never became binding precedent for other cases. Similar lawsuits from authors, publishers and record labels are still proceeding against Google, Meta, Midjourney and OpenAI, and each of those companies could be held to different standards depending on how it obtained its training data.

Still, the roughly $3,000-per-work rate could become a benchmark in other AI companies' settlement talks with rights holders, much as earlier music-industry settlements set rates for subsequent disputes in that field. For Anthropic itself, the $1.5 billion figure, while record-setting, represents only a fraction of the company's valuation, which funding rounds have placed in the tens of billions of dollars.

Polish Context

The case matters beyond the United States as well. Polish writers, translators and musicians have reported for months finding fragments of their work in the training data of large language models, without any real way to seek compensation under EU law, which, unlike US regulations, does not provide for similar class-action damages suits. The American settlement shows the scale of payouts courts can force once unlawful content acquisition is proven, but it gives creators in Europe no ready legal tool for equivalent claims.

For tech companies across the industry, this is a signal that legal costs tied to training data are no longer a theoretical risk. Further rulings and settlements in ongoing cases against other model providers may show in the coming quarters whether the rate set by Anthropic becomes a market norm or remains an exception tied to the specifics of this one case.