The lawsuit was filed in 2024, and the U.S. District Court for the Northern District of California recently finalized the settlement, according to Reuters.
According to the publication, this is the largest copyright infringement settlement in U.S. history.
Interestingly, in September 2025, the court ruled that training the model itself fell under the fair use doctrine and did not infringe on copyright. However, the court deemed it a violation for the company to store more than 7 million pirated copies of books in its own library, which theoretically could have been used for purposes other than training the AI. In other words, the problem lies specifically with the pirated copies of the books, not with the training of the language model itself.
After the court reached this conclusion, a settlement agreement was preliminarily approved that same month, in September 2025.
However, some authors opposed it. They claimed that the compensation amount was insufficient, that attorneys were receiving excessively high fees, and that some rights holders had been unjustifiably excluded from the agreement.
All of these objections were rejected by the San Francisco court on July 20.
Apana Shridhar, Deputy General Counsel at Anthropic, stated that more than 91% of the authors and publishers covered by the settlement had filed claims for compensation, and the company expects to finally close the case.
