A U.S. federal judge has granted final approval to Anthropic’s $1.5 billion settlement with authors and publishers, resolving the class action over books downloaded from pirate libraries. The agreement leaves intact an earlier ruling that Anthropic’s specific use of books to train Claude was fair use, although that ruling remains a nonbinding district court decision outside this case.
The settlement math
Judge Araceli Martínez-Olguín of the U.S. District Court for the Northern District of California approved the settlement on July 20, 2026. The company has not paid the full $1.5 billion at once: the official settlement FAQ says the fund is being financed through four instalments, beginning with $300 million paid in October 2025 and ending with a final payment due in September 2027. The final approval followed preliminary approval issued by now-retired Judge William Alsup in September 2025.
The court’s order identifies 482,460 works on the settlement list. Claims had been submitted for 440,490 of those works by April 16, representing 91.3% of the list, with the estimated payment remaining approximately $3,000 per work before costs, fees and any division among multiple rightsholders. Authors and publishers may therefore split the award for an individual title depending on their contracts and ownership rights.
A split ruling that favours the industry
Alsup’s June 2025 order drew a sharp distinction between training an AI model and acquiring the material used to build its library. He found that Anthropic’s use of the books at issue to train Claude was transformative and qualified as fair use, and that converting purchased print books into searchable digital copies was also permitted. But he rejected Anthropic’s argument that downloading and retaining millions of pirated books for a permanent, general-purpose library was protected by the same doctrine.
A damages trial over those pirated library copies was scheduled before the parties reached the settlement. Anthropic deputy general counsel Aparna Sridhar said the company regards the fair-use ruling as unchanged, while the settlement resolves the class claims connected to how the affected books were obtained and stored.
Why this is not the precedent it appears to be
The settlement ended the case before an appeals court could review Alsup’s fair-use holding. His decision therefore remains on the books, but it does not bind judges handling other copyright disputes involving AI companies. Another court could apply the four fair-use factors differently based on how training material was obtained, what the model produces and what evidence copyright owners present about market harm.
The ruling also covered the specific books and parties before Alsup rather than creating a blanket right for every AI developer to train on any copyrighted work. Its value to the industry is substantial, but its legal reach is narrower than the headline victory might suggest.
The structural read
For Anthropic, $1.5 billion is a significant settlement but not an existential financial blow. The company’s latest funding round valued it at $965 billion post-money, placing the settlement at a small fraction of that valuation. The agreement resolves past claims involving works on the approved list, but it does not function as a licence for future conduct or eliminate potential claims involving later activity and AI-generated outputs.
Rights holders can point to the largest known recovery in a U.S. copyright case, while AI developers can point to a surviving ruling that treated Anthropic’s training use as fair use. Both outcomes matter, but neither settles the wider legal fight. The money is real, and the fair-use decision remains intact, but a binding national rule for generative AI training still does not exist.
