Anthropic's $1.5bn book payout sets a price on pirated AI training data

Anthropic has put a dollar figure on the cost of training AI models with pirated books: $3,000 a title. On Monday, US District Judge Araceli Martínez-Olguín granted final approval to the company's $1.5 billion settlement with authors and publishers, closing the largest copyright case of its kind in US history and starting payments across roughly 500,000 works, according to TechCrunch.

The case turned on how Anthropic sourced the books it used to train Claude. The company bought and scanned physical copies of some titles, but also downloaded others from pirate sites. Courts had already ruled that training AI models on copyrighted text counts as fair use — but that pulling books from piracy sites does not. Facing trial on the piracy claims alone, Anthropic chose to settle.

Authors will now receive payments averaging about $3,000 per work, a figure that scales into the hundreds of millions once applied across the roughly 500,000 books covered by the deal. The settlement resolves Anthropic's exposure in this case but sets no binding precedent for the rest of the industry.

That leaves plenty still in play. Google, Meta, Midjourney and OpenAI all face pending suits over AI training data, and in July publisher Hachette and author Scott Turow sued Google separately over Gemini's training data. None of those cases are bound by the terms Anthropic just agreed.

The number that matters here is $3,000 a book, not $1.5 billion. It gives plaintiffs' lawyers a benchmark figure to demand in every other pending suit, and gives AI labs a rough price tag for the risk of using unlicensed text. Labs still relying on scraped or pirated corpora now have a concrete incentive to move towards purchased or licensed datasets before a court forces the issue.

Anthropic, meanwhile, buys itself certainty rather than vindication. Settling closes the piracy question in this case, but the fair-use finding on legitimately acquired books remains a going concern for the whole sector — one Google, Meta, Midjourney and OpenAI are all still fighting in parallel courts, with no guarantee they get Anthropic's price.