Anthropic’s $1.5 Billion Copyright Deal: Is the Record Settlement Just an R&D Expense?

The latest development in U.S. copyright law is that generative artificial intelligence (genAI) company Anthropic has agreed to settle a class-action lawsuit brought by a group of authors and publishers for a record $1.5 billion. The dispute centers on allegations that Anthropic used copyrighted works without permission to train its genAI model, Claude, and now has given rise to the largest settlement in U.S. copyright litigation history.

This case has raised key issues surrounding the intersection of artificial intelligence, copyright, and fair use, though the settlement will likely prevent a full judicial resolution of these issues, at least with respect to the particular circumstances of Anthropic’s case. Nonetheless, its resolution will likely set important precedents (or, at the very least, inform other judges) for future legal battles in this space.

The lawsuit stemmed from accusations that Anthropic had used a large volume of copyrighted materials, including books and written works, to train Claude. The plaintiffs argued that Anthropic did so without obtaining appropriate licenses and/or permission from the authors of the works. Anthropic, much like other genAI companies, responded to the allegations in part by claiming that their use of copyrighted works in genAI training was protected under the doctrine of fair use, which allows limited use of copyrighted material under certain conditions.

The proceedings took a pivotal turn in June 2025, when Judge William Alsup of the U.S. District Court for the Northern District of California ruled that the use of copyrighted content to train genAI models could qualify as fair use, but only in specific situations. Particularly, Judge Alsup found that using digital scans of legally purchased copies of books to train a genAI model fell within the bounds of fair use. However, he also found that Anthropic’s use of a dataset containing pirated e-books did not qualify as fair use. Indeed, Judge Alsup emphasized that retaining these pirated works for the purpose of building a general-purpose library for training violated the core principles of copyright law.

In the wake of the ruling, the case moved to a settlement, with Anthropic agreeing to pay $3,000 per work to over 500,000 authors and rights holders whose works had been used without authorization. This payment structure is part of the $1.5 billion settlement, which also includes a commitment by Anthropic to delete any pirated works from its systems. This settlement, which is still subject to final approval by Judge Alsup, will likely close one of the first major U.S. court cases to directly address the tension between genAI development and copyright protection, and will likely inform rulings in similar cases that are still making their way through the courts.

Thus, the significance of this case and its settlement may be broadly applicable. The ruling suggests that genAI developers can, in some cases, use copyrighted works without permission as long as the use is transformative and does not substitute for the original works. However, the retention of pirated works was clearly outside the bounds of fair use, highlighting the risks for companies that rely on such materials without securing proper licenses. Indeed, the “transformative” nature of the genAI’s usage of copyrighted material may prove to be key; in his June ruling, Judge Alsup described Claude’s such usage as “spectacularly transformative.” However, other similar lawsuits brought against genAI companies, such as that brought against OpenAI by the New York Times, center on allegations that genAI models can recite portions of their training data verbatim, which would undercut a claim of “transformative” use.

While the $1.5 billion sum is unprecedented, it represents only a fraction of what genAI companies like Anthropic are spending on model training and deployment, and what investors have committed to such companies. It is also far less than the potential statutory maximum of $150,000 per act of copyright infringement that Anthropic might have faced if the case had gone to trial. Also, the remuneration to individual authors is not significant enough to meaningfully compensate for the unauthorized use of their works or to alter their economic position. For that reason, the settlement may be perceived less as a remedy for authors than as a manageable expense for genAI companies. Given the scale of the industry, the outcome could reinforce a perception that large content licensing payouts are simply a negotiable cost of doing business, raising concerns about whether these companies will be deterred from future unauthorized use of copyrighted works.