A San Francisco federal judge granted final approval to a $1.5 billion settlement with Anthropic over its training of Claude on pirated books, marking the largest copyright payout in U.S. history and bringing closure to a landmark class-action brought by authors. The deal, which Anthropic says it does not admit wrongdoing on, requires payment into a fund covering over 480,000 works and orders the destruction of pirated files used in training. The settlement’s scope is narrow, covering how Anthropic acquired the books but not licensing ongoing use of the material for training or the outputs Claude generates, and authors retain the right to sue over outputs or any non-listed works. The case followed a split ruling by Judge William Alsup, who found fair use for training on the books but concluded that storing more than seven million pirated copies in a central library violated authors’ rights, and he has since retired. Since then, judges and lawyers have underscored that the settlement will influence how AI developers assess legal exposure around training data, even as some authors argue the amount is insufficient or overly favors counsel. The agreement is one of many high-profile copyright lawsuits against AI firms, and it underscores ongoing tensions between innovation, fair use, and compensation for authors in the age of large language models.