The Anthropic Copyright Settlement: What Happened and Why It Matters

The Lawsuit

In 2024, authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson filed a class-action lawsuit against Anthropic, alleging that the company had downloaded millions of copyrighted books from so-called shadow libraries — pirated online repositories such as Library Genesis and Pirate Library Mirror — to help train its Claude AI models. The case, formally known as Bartz v. Anthropic, became one of the most closely watched legal proceedings testing how copyright law applies to AI training.

A Split Ruling on Fair Use

In June 2025, U.S. District Judge William Alsup issued a landmark ruling that split the underlying legal question in two. He found that using copyrighted books to train a large language model could qualify as fair use, but only if the books had been acquired legally in the first place. Acquiring books through pirated shadow libraries, by contrast, did not qualify as fair use, and Alsup ordered that question to proceed to a separate trial to determine Anthropic's liability and the resulting damages.

The Scale of the Risk

The stakes of that trial were enormous. U.S. copyright law allows for statutory damages of up to $150,000 per willfully infringed work, and the court's own findings indicated Anthropic had downloaded more than seven million pirated book copies. With a class eventually certified covering roughly 482,000 individual works that met specific registration criteria, Anthropic faced theoretical liability that some estimates placed well above $70 billion had the case gone to a full trial and resulted in maximum statutory damages.

The Settlement

Facing that exposure, Anthropic agreed in late August 2025 to settle the piracy claims for $1.5 billion plus interest, in what was described by multiple parties as the largest copyright class-action settlement in U.S. history. Under the settlement, eligible rightsholders — a group that includes both authors and the publishers who hold reproduction rights to a given book — were set to receive roughly $3,000 per qualifying work, a figure that could rise depending on how many eligible claimants ultimately filed. As part of the deal, Anthropic also agreed to destroy the pirated datasets and any derivative copies within thirty days of final judgment and to certify that destruction to the plaintiffs' counsel.

What the Settlement Did and Didn't Resolve

It is worth being precise about what the settlement actually covers. It resolves Anthropic's legacy liability for having acquired books through piracy — the narrower question the court had already found problematic. It does not overturn, and in fact builds on, the broader June 2025 ruling that training an AI model on legally acquired copyrighted books can itself qualify as fair use. Some legal commentators have described the outcome as effectively establishing, at least within that court's jurisdiction, that AI companies can train on copyrighted material so long as they obtain it through legitimate channels rather than piracy.

Why the Case Reverberated Across the AI Industry

Because dozens of similar copyright lawsuits remain pending against other major AI companies, the size and structure of the Anthropic settlement was widely interpreted as setting an informal benchmark for how such cases might eventually be resolved elsewhere. Legal experts noted that while $1.5 billion was manageable for a company of Anthropic's scale and valuation, a similar per-work damages figure could pose a much more serious threat to smaller AI companies facing comparable claims, and the case has added momentum to a broader industry shift toward licensed, permission-based data deals for AI training going forward.