A federal judge approves Anthropic's $1.5 billion copyright settlement with authors and publishers
The payout works out to $3,000 per book across roughly 500,000 works. But the ruling that made AI training on copyrighted text legal still stands, and many authors are not celebrating.
Key points
- A federal judge gave final approval on 21 July 2026 to Anthropic's $1.5 billion settlement of a copyright class action lawsuit brought by authors and book publishers.
- The settlement covers an estimated 500,000 works, paying $3,000 per work to the authors and publishers who hold rights to them.
- An earlier ruling found that training an AI model on copyrighted text counts as "fair use" under US law, a finding that favours the AI industry broadly.
- Anthropic settled to avoid a jury trial over its use of pirate download sites to build its training library.
- The settlement closes this case but does not create binding law; copyright lawsuits against Google, Meta, Midjourney, and OpenAI are still active.
For authors who sued Anthropic, the long legal fight is over. On Monday, Judge Araceli Martinez-Olguin of the US District Court for the Northern District of California signed off on a $1.5 billion settlement between Anthropic, the company behind the Claude AI assistant, and a class of authors and publishers whose books were used to train it. TechCrunch AI first reported the approval.
The money is substantial. Roughly 500,000 books are covered, and rights-holders will receive $3,000 per work. Lawyers and legal experts describe this as the largest copyright settlement in US history.
So why are many writers still unhappy?
The answer is in the ruling that preceded the settlement. The original judge on the case, William Alsup, decided the core legal question in Anthropic's favour. He ruled that using copyrighted text to train an AI model, meaning feeding it millions of books so it learns patterns of language and fact, counts as "fair use." Fair use is a legal doctrine that allows limited use of copyrighted material without permission, typically for purposes like education or commentary. Applying it to AI training is a big deal. It means the act of building an AI on someone's book is, under this reading, not itself a violation.
What Alsup did find illegal was how Anthropic got hold of some of those books. The company built its training library two ways: buying and scanning books (permitted), and downloading books from piracy sites like Library Genesis and Pirate Library Mirror (not permitted). That second route was the problem. Alsup said the piracy question could go before a jury. Anthropic settled rather than face that risk.
The settlement ends this specific case. But it does not settle the wider legal argument, and that matters for every author, publisher, and AI company involved in the dozen-plus copyright cases still working through US courts.
Alsup's fair-use ruling was a single district court decision. Because Anthropic settled, the case will never reach a federal appeals court, which is the level where rulings become binding on other judges. Every other judge hearing a similar case can still reach a different conclusion.
That fight is very much alive. Just last week, a group including publishers Hachette and Elsevier, and author Scott Turow, filed a new class action lawsuit against Google, accusing the company of using their copyrighted works to train Gemini, Google's AI assistant.
What does this mean for authors whose books are used to train AI?
Right now, US law on this question is genuinely unsettled. The Anthropic case produced one district court's view that AI training is fair use, but that view is not binding on other courts. Authors whose books appear in AI training data do not yet have a clear legal right to compensation, and the outcome of the Google, Meta, and OpenAI cases could shift that picture significantly in either direction.
