It is a figure that will go down in copyright history in the age of AI. The American judiciary has given its final green light to the agreement reached between Anthropic and tens of thousands of authors whose works had been pilfered to train the company’s model.
The case originated from a complaint filed by three American writers, thriller novelist Andrea Bartz and non-fiction authors Charles Graeber and Kirk Wallace Johnson, who accused Anthropic of scraping their works without permission.
Their action escalated into a class action ultimately gathering more than 500,000 titles. Under the terms of the settlement, each affected book would entitle the author or their publisher to compensation of about $3,000.
The plaintiffs’ lawyer hailed what he called the largest ever compensation obtained in an intellectual property case, a first for the AI industry.
The federal judge William Alsup, who oversaw the case in San Francisco before retiring, delivered in June a ruling with two faces.
He held that training an AI model on copyrighted books did not, in itself, constitute an infringement, falling under the principle of “fair use.” However, he ruled that Anthropic had obtained these works illegally, by tapping into pirate digital libraries.
A $3,000-per-book settlement
More than 7 million digitized books had thus been downloaded with full knowledge of their fraudulent origin. Initially nearly 200,000 volumes from a database called Books3, then at least 5 million copies retrieved from the pirate site Library Genesis (LibGen) and 2 million additional via the Pirate Library Mirror. Andrea Bartz’s novel “The Lost Night” was among the works found in these datasets.
Without this amicable settlement, Anthropic faced a trial slated for December, the outcome of which could have cost the company much more. A legal analyst from Wolters Kluwer noted the prospect of a multi-billion-dollar bill, capable of endangering the company’s very viability.
As part of the agreement, Anthropic also agreed to destroy the original files of the books that had been illegally downloaded.
As a sign of the case’s magnitude, of the approximately 482,000 books covered by the final decision, 91% have already been claimed by their authors or publishers, now awaiting settlement.
It was Judge Araceli Martínez-Olguín, who took over the case after Alsup’s retirement, who endorsed the agreement, ruling that it provided substantial redress to the harmed authors.
Authors Guild, the organization that defends the interests of American writers, praised the outcome as a model result for authors and publishers, saying it sends a strong signal to the AI industry about the consequences of pirating protected works.
This case is only the first in a long series: dozens of similar lawsuits against other AI giants are still pending before the American courts.
