Humanity has agreed to pay $1.5 billion to resolve lawsuits filed by authors and publishers over the use of millions of copyrighted books to train the AI chatbot Claude model, according to legal filings posted online.
A federal judge discovered in June that while humanity's use of seven million pirated books is fairly protected, retaining digital works in the “Central Library” violates copyright law. The judge determined that the company's executives knew they were downloading the pirated work, and found that a trial was scheduled for December.
The settlement presented to a federal judge on Friday remains with final approval, but it will pay hundreds of thousands of authors $3,000 per book, according to the New York Times. The $1.5 billion settlement is the largest payment in the history of US copyright law, but the amount paid per job is often higher. For example, in 2012, a Minnesota woman paid about $9,000 per song. This is the number that was defeated after being initially ordered to pay more than $60,000 per song.
In a statement to Gizmod on Friday, humanity touted its previous ruling from June as engaged in fair use by training models on millions of books.
“In June, the district court issued a groundbreaking ruling on AI development and copyright law, finding that humanity's approach to training AI models constitutes fair use,” Aparna Sridhar, human assistant adviser, said in an email statement.
“Today's settlement, if approved, will settle the plaintiff's remaining estate claims. We are committed to developing secure AI systems that help people and organizations expand their capabilities, advance scientific discoveries and solve complex problems,” Sridhar continued.
According to legal submissions, humanity says that payments will be made. With four tranches tied to court-approved milestones. The initial payment will be $300 million within five days of preliminary approval of the court's settlement and another $300 million within five days of the final approval order. $450 million will then be held in interest within 12 months of the reserve order. And finally, within the next year it was $450 million.
Humanity, which was recently valued at $183 billion, signed a deal in early 2024 to train AI models with content on the platform. The author has aggressively filed lawsuits against other large tech companies such as Openai, Microsoft, and Meta.
The ruling from June explained that humanity's training in AI models using copyrighted books is believed to be used fairly under US copyright law.
…It is not a creative element of a particular work, and even the author's identifiable, expressive style is not generally reproduced. Yes, Claude outputs grammar, composition and style that the underlying LLM has distilled from thousands of works. But if someone reads all modern classics for extraordinary expressions, remember them, and then emulates their best blend of writing, would it violate copyright law? Of course it's not.
“Like readers who are aiming to become writers, the LLMS of humanity trained their works to not race or replicate or take them first, but to turn hard corners to create something different,” the verdict said.
Under this legal theory, the company was to buy all the books that became pirated to legally train all models. However, as the New York Times notes, the settlement cannot go to trial and therefore does not set legal precedents that can determine future cases.
