Anthropic just paid $1.5 billion for pirated books and still won
Anthropic will pay $1.5 billion to authors after downloading their books from pirate sites to train Claude. A federal judge in San Francisco granted final approval on July 20, closing what plaintiff attorneys call the largest copyright recovery in history.
Here is the part nobody is highlighting enough: Anthropic actually won the legal argument that matters.
Judge William Alsup, who issued the original split ruling before retiring, determined that training an AI model on copyrighted books constitutes fair use under US copyright law. The $1.5 billion is not a penalty for training Claude on copyrighted material. It is a penalty for how Anthropic acquired those books in the first place.
Anthropic downloaded millions of books from Library Genesis and Pirate Library Mirror instead of buying them. That is what cost $1.5 billion. The training itself was ruled legal.
Anthropic’s deputy general counsel Aparna Sridhar wasted no time framing the ruling as a victory, calling it proof “that training AI on books is fair use under copyright law.” For a company that just agreed to pay $1.5 billion, that is a remarkably aggressive spin. It also happens to be accurate.
The numbers
The settlement covers over 482,000 books. Authors receive roughly $3,000 per title, which is about four times the usual minimum statutory damage award for copyright infringement. 91 percent of eligible works have already been claimed by their authors or publishers.
Attorney fees were originally set at 12.5 percent of the settlement, or $187.5 million. The court cut that to approximately 6.8 percent, around $101 million.
Anthropic is also required to destroy all pirated files obtained from Library Genesis and Pirate Library Mirror.
What this means for every other AI company
The precedent here is sharp. Training on copyrighted works: legal. Acquiring those works through piracy: expensive.
Every AI lab facing similar lawsuits just got a roadmap. The Bartz v. Anthropic ruling separates data acquisition from data use, and only the acquisition carries liability. OpenAI, Meta, and others defending copyright lawsuits will cite this case to argue that their training process itself is protected. Whether they downloaded books from pirate sites, scraped them, or licensed them will determine their exposure.
The settlement does not release Anthropic from future claims. Authors who opted out, including Nolan Bushnell and roughly 100 others, have filed separate lawsuits that are still working through the courts. The agreement also explicitly preserves the right to sue over what AI models generate, not just what they were trained on.
Thriller novelist Andrea Bartz filed the original suit in 2024 alongside Kirk Wallace Johnson and Charles Graeber. Judge Araceli Martínez-Olguín took over after Alsup’s retirement and signed the final order, calling the settlement “meaningful relief” for affected authors.
This is the first major settlement in a wave of AI copyright cases still pending across federal courts. It will not be the last. But the legal framework it establishes, that training is fair use and piracy is piracy, gives AI companies something they have wanted since these lawsuits started: a clear line between what they do with data and how they get it.
The line just cost Anthropic $1.5 billion to draw.