A federal court in San Francisco has finally approved a settlement agreement under which Anthropic will pay $1.5 billion to authors for the unauthorized use of their books in training its language model, Claude. This is the largest known copyright settlement in U.S. history.

District Judge Araceli Martinez-Olguin overruled all objections, including claims that the payout amount was insufficient. The agreement was reached back in 2025 following a preliminary ruling by Judge William Alsup, who established the dual nature of the dispute: training AI on copyrighted books may constitute fair use, but storing over seven million pirated copies in a "central library" does not.

The lawsuit was filed in 2024. The authors alleged that Anthropic used pirated versions of their works without permission. Aparna Shridhar, the company's deputy general counsel, confirmed that over 91% of authors and publishers covered by the deal have already requested their share of the payments.

Lead attorney for the plaintiffs, Justin Nelson, called the agreement historic. The judge also awarded the lawyers over $101 million of the requested $187.5 million, reducing the fee amount. Some authors and publishers opted out of the agreement and filed separate lawsuits against Anthropic — those proceedings are ongoing.

My analysis: This ruling sets a crucial precedent for the entire artificial intelligence industry. It clearly separates two aspects: using content for training (permissible) and storing pirated copies (impermissible). But the main lesson here is the cost of a mistake. $1.5 billion is not just compensation; it's a signal to the market: saving on data licensing can end up costing more than any expenses for legal datasets. In the coming years, we will see a massive overhaul of approaches to collecting training data among all major AI companies.