Copyright lawyers got good news earlier this week, in the form of a settlement decision that could lead to even more lawsuits against AI companies.
On Monday, Judge Araceli Martínez-Olguín of the U.S. District Court for the Northern District of California approved a $1.5 billion settlement agreement between AI company Anthropic and a group of authors who accused it of using their books to train its large language models without their permission.
It’s one of the largest copyright settlements in history, according to the authors’ attorney, Justin Nelson, and could be precedent-setting for pending copyright suits against AI companies.
The settlement is the culmination of a nearly two-year legal process that began in August 2024, when nonfiction authors Charles Graeber, Kirk Wallace Johnson, and Andrea Bartz sued Anthropic under the Copyright Act of 1976, alleging the AI giant engaged in “largescale theft of copyrighted works.” Their complaint was later amended to ensure the suing authors had standing, but the accusations remained the same.
In June 2025, District Court Judge William Alsup (who retired in December 2025) ruled partly in Anthropic’s favor, finding that the company’s use of legally acquired books to train its model constituted “fair use under Section 107 of the Copyright Act.” Under the law, works used for “criticism, comment, news reporting, teaching (including multiple copies for classroom use), scholarship, or research” are considered fair use.
In his 2025 ruling, Alsup agreed with Anthropic’s argument that there was nothing illegal about converting physical copies of books it purchased into digital versions for training. However, he did find that the company’s use of pirated books was “not justified by a fair use,” and he refused to grant Anthropic’s request to shield its repository of copyrighted works “not used for training” from future litigation.
Both sides agreed to the terms of the settlement after prolonged negotiations with a third-party mediator. Alsup issued preliminary approval of the settlement terms in September 2025. Authors and publishers had until March 30 to join the claim or opt out.
Under the terms of the agreement, members of the class action suit who declined to opt out of the settlement “released and forever discharged” any claims of copyright infringement if their works were one of the nearly 500,000 books that met the class certification requirements. The settlement doesn’t cover “future conduct” by Anthropic, and it doesn’t prevent authors whose books were used by Anthropic but not on the court-approved list of pirated works from suing the company for copyright infringement.
The $1.5 billion award will be split among authors and publishers after fees and expenses are paid, netting everyone entitled to a claim roughly $3,000 per work, “four times the value of the statutory minimum,” according to Nelson.
Anthropic’s decision to settle the case might hurt the company’s pocketbook for now. Still, it likely avoids a court battle with even wider-reaching implications, since many AI companies use pirated data to train their models. Indeed, AI leaders such as Meta, Perplexity AI, Google, OpenAI, and Nvidia are facing lawsuits over similar copyright infringement claims. Anthropic is also being sued by Concord Music Group, which alleges that the company used copyrighted lyrics to train its Frontier AI model, Claude.
Time will tell how these disputes shake out, but courts placing restrictions on fair use of publicly available data will likely make AI models worse. Instead of settling potential copyright claims, AI companies will likely turn to a growing market for repositories of licensed copyrighted works.
Of course, this will also make production of the models more expensive, a cost that will inevitably be passed on to consumers.
