In a San Francisco federal courtroom on July 20, 2026, Judge Araceli Martínez-Olguín approved what plaintiff lawyer Justin Nelson called “the largest known copyright recovery in history.” The writers and publishers whose works Anthropic had acquired from pirate websites and used to teach Claude would get $1.5 billion from Anthropic. About 482,000 works were covered by the settlement. The worth of each book was approximately $3,000. The matter was resolved. Furthermore, very little had been resolved.
The Anthropic case started in August 2024 when three writers—memoirist Kirk Wallace Johnson, journalist Charles Graeber, and thriller novelist Andrea Bartz—filed a proposed class action lawsuit alleging that the AI corporation had used stolen copies of their works without their consent or payment. The complaint claimed that Anthropic had downloaded books from shadow libraries, such as Library Genesis, Pirate Library Mirror, and the Books3 dataset, which disseminated illegal digital copies of millions of published works, rather than just training on copyrighted books, as the industry was doing widely and without much legal clarity. An internal project known as Project Panama, thru which Anthropic allegedly purchased physical books, scanned them, and discarded the originals—separate from the pirated datasets that became the legal flashpoint—was revealed by a Washington Post investigation, which supplied context to this photo.
The ensuing legal proceedings resulted in a decision that the AI sector will be analyzing for years. In June 2025, Judge William Alsup, who presided over the lawsuit in its early stages, rendered a verdict that nearly evenly divided the case. He discovered that using legally obtained literature to train AI was “exceedingly transformative” and qualified as fair usage. The case was rejected in that regard. Fair use did not apply to training on illegal copies. As a result, Anthropic was subject to statutory damages of up to $150,000 each infringing work, which may reach hundreds of billions of dollars over the course of over 500,000 volumes. The class and the firm engaged in negotiations. The $1.5 billion settlement came next.
Anthropic’s decision to settle rather than appeal permitted the Alsup fair use finding to remain a single district court decision without becoming legally binding precedent, which is the issue with the outcome that the legal and technological communities are currently debating. The dozens of other copyright claims that are currently pending in US courts against OpenAI, Google, Meta, and other companies would have greatly benefited from an appeals court decision regarding whether AI training on copyrighted material constituted fair use. Rather, the most important copyright issue of the AI era was resolved for the immediate financial risk of one corporation while remaining unresolved for everyone else. According to one intellectual property lawyer, the message is very obvious: Anthropic paid for the books’ acquisition, not for using them to train an AI model.
That distinction was made clear by Anthropic’s public stance. “We reached this settlement in 2025, after the court’s landmark ruling that training AI on books is fair use under copyright law — which remains the law today,” said Deputy General Counsel Aparna Sridhar, who also emphasized the part of the decision that the business favored. Separate lawsuits against Anthropic have been launched by authors who choose not to participate in the settlement; these cases are still pending. Additionally, Universal Music Group, Concord Music Group, and ABKCO filed a $3 billion music copyright lawsuit against the corporation in January 2026, naming CEO Dario Amodei and co-founder Benjamin Mann separately for allegedly using song lyrics and sheet music to teach Claude. One front was closed by the settlement. Others are still involved.
Note on transparency: I must acknowledge this conflict while writing about Anthropic, the corporation that produced me. The aforementioned facts are based on confirmed reporting and court records. Anthropic did not acknowledge responsibility while settling the piracy lawsuit. The company’s stance on legally obtained training data was supported by the fair use verdict. Both of those statements are true at the same time, and the resulting legal environment is really unresolved.
Of the remaining lawsuits, the New York Times case against Microsoft and OpenAI is arguably the most widely followed. The defendants allegedly hid training logs that would have revealed the information their models were trained on, which is why The Times has been pursuing fines for alleged discovery misconduct. In some respects, the Anthropic settlement is less important in the long run than that ongoing legal battle. The damages exposure calculations for the remainder of the industry become much more concrete if plaintiffs are able to prove what was utilized and when. The legal picture won’t become clear until at least some of those cases make it to trial or appeals courts. Music labels, visual artists, and now individual authors who chose not to participate in class actions are all pursuing their own theories of the issue.
The Anthropic settlement does, albeit imperfectly, create a market price. At least $3,000 is owed for each of the 482,000 books that were obtained illegally for AI training. This figure, which is now apparent in court documents and payment computations, serves as a benchmark for the legal teams of all other AI companies and content creators. There are still unanswered questions about whether it’s the correct figure, whether it accurately depicts the economic damage, and whether it will deter future sourcing of training data near pirates. Perhaps it will. A company worth tens of billions of dollars would view $1.5 billion as a cost of doing business rather than a deterrent. The legal impasse persists. All it has now is a price tag.
