Book authors filed a class-action lawsuit against Anthropic, an artificial intelligence business, alleging that the company used unlicensed versions of their books to train its chatbot. Anthropic has agreed to pay $1.5 billion to resolve the claim.
If a judge approves the historic deal as early as Monday, it might be a watershed in the legal disputes between AI corporations and the authors, visual artists, and other creative people who claim they violate copyright.
The deal covers an estimated 500,000 publications, and the business has agreed to pay authors roughly $3,000 for each book.
Justin Nelson, an attorney for the authors, stated, “As far as we can tell, it’s the largest copyright recovery ever.”
“In the AI era, it is the first of its kind.”
Three authors—nonfiction authors Kirk Wallace Johnson and Charles Graeber and thriller novelist Andrea Bartz—who filed a lawsuit last year now stand in for a larger number of authors and publishers whose works Anthropic used to train its chatbot Claude.
In June, a federal judge rendered a divided decision in the case, concluding that Anthropic had improperly obtained millions of books through piracy networks but that it was permissible to train AI chatbots on copyrighted works.
According to analysts, Anthropic might have lost the case during a planned trial in December and would have had to pay much more money if it hadn’t settled.
William Long, a legal analyst for Wolters Kluwer, stated, “We were looking at a strong possibility of multiple billions of dollars, enough to potentially cripple or even put Anthropic out of business.”
San Francisco-based US District Judge William Alsup has set a hearing for Monday to go over the terms of the settlement.
What role do books have in AI?
The AI large language models that power chatbots like Anthropic’s Claude and its main competitor, OpenAI’s ChatGPT, are built using data from books, which are essentially billions of words meticulously put together.
According to Alsup’s June verdict, Anthropic downloaded over 7 million digital books that it “knew had been pirated.”
In order to match the extensive collections that ChatGPT was trained on, it began with almost 200,000 from an online library called Books3, which was put together by AI researchers outside of OpenAI.
The Books3 dataset included the first thriller book, The Lost Night, written by Bartz, one of the case’s principal plaintiffs.
According to Alsup, Anthropic later stole at least 2 million copies from the Pirate Library Mirror and at least 5 million copies from the pirate website Library Genesis, also known as LibGen.
Last month, the Authors Guild informed its thousands of members that if Anthropic was found guilty of knowingly violating their copyrights at trial, “damages will be at least $750 per work and could be much higher.”
The increased settlement award of about $3,000 per work likely reflects a smaller pool of impacted books after copyright infringements and duplicates have been removed.
The CEO of the Authors Guild, Mary Rasenberger, referred to the settlement on Friday as “an excellent result for authors, publishers, and rightsholders generally, sending a strong message to the AI industry that there are serious consequences when they pirate authors’ works to train their AI, robbing those least able to afford it.”
