A U.S. judge has given final approval to a $1.5 billion settlement between AI company Anthropic and thousands of authors whose copyrighted books were used to train the company's AI without their permission. Judge Araceli Martínez-Olguín signed the order on Monday, July 20, 2026. Reuters first reported the story. Reuters

AI models learn by processing huge amounts of text. In Anthropic's case, that text included thousands of copyrighted books, used without the authors' consent. The settlement gives approximately $3,000 for each book Anthropic used in training. More than 91% of eligible authors and publishers claimed their share, according to Anthropic general counsel Aparna Sridhar. The Verge

The plaintiffs' law firm, Susman Godfrey, called the $1.5 billion figure the "largest known copyright recovery in history." The Verge

The lawsuit was first filed in 2024 by authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson. The case took a complicated path to resolution. In June 2025, a federal judge ruled that training AI on copyrighted books was not automatically illegal, but allowed other claims against Anthropic to continue. Judge William Alsup, now retired, granted preliminary approval to the settlement in September 2025. He had previously given Anthropic a partial win in the case, while also allowing a separate claim that the company engaged in "Napster-style downloading of millions of works." The Verge

Getting from preliminary to final approval was not straightforward. On September 9, 2025, a judge in San Francisco initially declined to approve the deal immediately. A fairness hearing took place on May 14, 2026, before Judge Martínez-Olguín, who signed the final approval order on July 20. Reuters

The settlement does not end all of Anthropic's legal trouble. The company still faces separate lawsuits from Chicken Soup for the Soul and other authors who argue that $3,000 per book is not enough. Those plaintiffs are pursuing their own claims outside this settlement. The Verge

The broader context here is what this deal means for the wider world of AI and copyright. The settlement puts a specific dollar amount, roughly $3,000, on each book used without permission to train an AI system. Whether that figure becomes a standard that other courts and companies adopt is an open question, but lawyers in future cases will certainly point to it as a reference. The high participation rate, over 91%, suggests most affected authors found the amount acceptable. The holdouts clearly disagree.

The case history also matters for other lawsuits against AI companies. The June 2025 ruling that training AI on copyrighted books is not automatically illegal, combined with other claims that survived, means courts are drawing lines: some uses of copyrighted material may be permitted under fair use rules, while others are not. The "Napster-style downloading" claim that survived separately suggests that how a company gets the books in the first place can be its own legal problem, separate from whether using them for training is allowed.

For AI companies still building or defending their training data, the Anthropic settlement shows what a resolution looks like in practice: a very large payment, broad participation from affected authors, and ongoing lawsuits from those who opted out or were not covered. The remaining cases will test whether $3,000 per book is the most an author can expect, or just the starting point.