A federal judge has granted final approval to Anthropic's $1.5 billion class action settlement with thousands of authors whose copyrighted books were used to train the company's AI models without permission. U.S. District Judge Araceli Martínez-Olguín signed the order on Monday, July 20, 2026. Reuters first reported the decision. Reuters

The settlement provides approximately $3,000 for each book Anthropic allegedly used in training, and covers thousands of authors and publishers. More than 91% of eligible authors and publishers claimed their share, according to Anthropic general counsel Aparna Sridhar. The Verge

Plaintiffs' law firm Susman Godfrey described the $1.5 billion figure as the "largest known copyright recovery in history." The Verge

The class action was originally filed in 2024 by authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson. The case took a winding path to resolution. In June 2025, a federal judge issued a mixed ruling: training AI chatbots on copyrighted books was not inherently illegal, but other claims against Anthropic survived. Then-presiding Judge William Alsup, now retired, granted preliminary approval to the settlement in September 2025. He had previously handed Anthropic a partial victory before allowing a separate class action alleging the company engaged in "Napster-style downloading of millions of works." The Verge

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

The settlement does not close out Anthropic's copyright exposure. The company still faces separate lawsuits from Chicken Soup for the Soul and other authors who argue that $3,000 per book is insufficient. Those plaintiffs are pursuing their own claims outside the class action framework. The Verge

The broader context here is what this settlement means for the wider landscape of AI training and copyright. The deal establishes a concrete per-book valuation for unauthorized use of books in model training. Whether $3,000 per book becomes a de facto benchmark is an open question, but opposing counsel in remaining and future cases will inevitably cite the figure as a reference point. The fact that more than 91% of the class participated suggests the per-book amount was broadly acceptable to affected rights holders, though the holdouts and separate plaintiffs clearly believe the number undervalues their work.

The procedural history also matters for parties litigating similar claims against other AI developers. The June 2025 ruling that training on copyrighted books was not illegal in itself, combined with the survival of other claims, sketches a legal terrain where fair use arguments for training can succeed on some dimensions while failing on others. The "Napster-style downloading" framing that survived as a separate claim points to how the books were acquired as a distinct liability issue, separate from the question of whether training itself infringes.

For AI companies still building or defending training pipelines, the Anthropic settlement offers a concrete data point on what resolution looks like in practice: a large dollar figure, broad class participation, and continued litigation from those who opt out or were not covered. The remaining lawsuits from Chicken Soup for the Soul and other authors will test whether $3,000 per book holds as a ceiling or becomes a floor.