A federal judge has approved Anthropic's $1.5 billion settlement with authors whose books were used to train its AI models. The agreement resolves claims of copyright infringement and marks the largest such settlement in U.S. history.
The settlement, approved by U.S. District Judge Araceli Martínez-Olguín, resolves a class-action lawsuit filed by authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson. The authors accused Anthropic of using pirated books downloaded from online libraries like LibGen and PiLiMi to train its Claude AI system. While a previous ruling by Judge William Alsup suggested that using lawfully acquired books for AI training could be considered fair use, he indicated that the acquisition and retention of pirated works were likely infringing. This distinction led to the settlement, which provides "meaningful relief" to authors, with payments estimated at approximately $3,000 per eligible book.
The lawsuit, initially filed in August 2024, represented a significant test of copyright law in the age of artificial intelligence. Anthropic had argued that its use of copyrighted material for AI training fell under fair use provisions. Judge Alsup's summary judgment ruling in June 2025 partially supported this, stating that the training use of legally obtained books was "quintessentially transformative". However, Alsup also found that Anthropic's practice of maintaining a permanent library of pirated books was not transformative and likely infringed copyright. This ruling set the stage for a trial on damages, which Anthropic sought to avoid by entering settlement negotiations.
The final settlement, preliminarily approved in September 2025, encompasses nearly 500,000 eligible works. The $1.5 billion award will be distributed among the rights holders of these books, including authors and publishers, after administrative and legal fees are deducted. Plaintiffs' attorneys were awarded $101.6 million in fees, a reduction from their initial request of $187.5 million. Anthropic is scheduled to pay the settlement amount in four installments, with the final payment due by September 2027.
The settlement is considered a landmark event, establishing a significant precedent for copyright disputes involving AI development. It is the first major settlement in a wave of copyright cases brought against AI companies over their training data. While Anthropic's use of lawfully acquired books for training is now largely settled as fair use, the question of acquisition and retention of pirated materials remains a critical point of contention, as highlighted by this case. The court's approval signifies that while AI training may be protected, the methods of data acquisition are subject to copyright law.
The implications of this settlement extend beyond Anthropic, potentially influencing how other AI developers approach the use of copyrighted material. Dozens of similar lawsuits are pending against other major AI companies, including OpenAI, Google, and Meta, involving various forms of creative content such as news articles and movies. The Bartz v. Anthropic case demonstrates the financial risks AI companies face if they do not secure proper licensing for training data, even if the training process itself is deemed fair use. The settlement also includes provisions for Anthropic to delete the pirated files it downloaded, further addressing the infringement claims.