The AI copyright debate has produced its largest financial resolution yet – but not a definitive rule for the entire industry.
A federal judge has granted final approval to Anthropic’s $1.5 billion settlement with a class of authors and publishers who alleged the company illegally obtained copyrighted books while building its AI training library. Judge Araceli Martinez-Olguin approved the agreement after retired Judge William Alsup issued preliminary approval in 2025.
The settlement is expected to provide approximately $3,000 per eligible work across an estimated 500,000 books, with payments shared among the authors and publishers who hold the relevant rights. It is believed to be the largest monetary recovery in US copyright history.
The dispute was about acquisition, not only training
The case involved two separate questions: whether Anthropic could use copyrighted books to train an AI model, and whether it had lawfully obtained those books.
Alsup ruled that using legally acquired books for AI training could qualify as transformative fair use. He also accepted Anthropic’s practice of purchasing physical books and digitizing them for internal use.
The court reached a different conclusion regarding books downloaded from pirate repositories, including Library Genesis and Pirate Library Mirror. Alsup found that obtaining and storing those unauthorized copies was independently unlawful, even if the broader training activity could qualify as fair use.
Anthropic agreed to settle before the piracy claims reached trial, avoiding the possibility of substantially higher statutory damages.
Why authors remain divided
The approval gives eligible rights holders a path to compensation, but some authors have criticized the claims process and the amount ultimately available per title.
The settlement also does not establish that AI companies must license every copyrighted work used for training. Anthropic’s earlier fair-use victory remains intact within this district-court case, while the payment resolves the narrower issue of how the books were acquired.
That distinction explains why both sides can present parts of the outcome as a victory. Authors and publishers secured a record financial settlement over unauthorized copying, while Anthropic preserved a favorable ruling on transformative AI training.
The market signal
The case creates a clearer compliance warning for AI developers: the source of training data can matter as much as the purpose for which it is used.
Companies may strengthen provenance checks, licensing arrangements and internal documentation around training datasets. Publishers, meanwhile, have stronger evidence that unauthorized acquisition can create major financial exposure even when model training itself receives fair-use protection.
The decision does not bind every US court. Because Anthropic settled rather than appealing the case, other judges remain free to reach different conclusions in lawsuits involving Google, Meta, Midjourney, OpenAI and other AI companies.
The settlement closes one landmark case, but the larger legal market remains unsettled. The next phase will focus on whether AI developers can prove not only that their training is transformative, but that the data entering their systems was obtained lawfully.
Source : TechCrunch
