Anthropic just agreed to fork over at least $1.5 billion to authors whose books were allegedly used to train its AI systems.
That breaks down to about $3,000 per book, making this possibly the first time in history that pirated PDFs have paid more than publishing royalties.
The deal, revealed in a court filing this week, is being called the largest publicly reported recovery in US copyright litigation.
It’s not a done deal yet. A judge still needs to sign off, with a hearing scheduled for September 8. But if approved, around 500,000 works are expected to be covered.
And if even more authors line up with claims, Anthropic will keep paying out $3,000 a pop until the checks (plus interest) are mailed.
Oh, and the company has to destroy all the pirated files it downloaded.
The fine print matters: the settlement only covers past acts. Anthropic doesn’t magically get a license for future training, and any new claims after August 25, 2025, are still fair game.
This showdown started back in August 2024, when authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson accused Anthropic of building a multibillion-dollar AI business on stolen books.
Judges handed Anthropic a partial win, saying training on legally purchased works is fair game, but the pirated book issue went to trial.
By July, the case ballooned into a class action on behalf of US writers. Fast forward to now, and Anthropic is basically paying to close the book on the matter.
Anthropic says it’s happy to move on. “Today’s settlement, if approved, will resolve the plaintiffs’ remaining legacy claims,” deputy general counsel Aparna Sridhar told The Verge, while promising the company still wants to build “safe AI systems.”
The payout lands at a pivotal moment: lawsuits against AI firms are piling up, while media outlets and platforms are also striking data deals with those same companies.
Anthropic itself is still facing cases from Reddit and Universal Music over alleged misuse of posts and song lyrics.
Does Anthropic’s $1.5 billion settlement set a fair precedent for AI companies using copyrighted content, or will these massive payouts stifle AI development and innovation? Should authors be compensated every time their work helps train AI systems, or is this just the cost of doing business in the digital age? Tell us below in the comments, or reach us via our Twitter or Facebook.
