A federal judge on Monday gave final approval to Anthropic's $1.5 billion settlement with a class of authors and book publishers, clearing the way for the largest copyright payout in U.S. history. The deal delivers $3,000 per work across an estimated 500,000 works, shared among the rights holders. Judge Araceli Martinez-Olguin of the U.S. District Court for the Northern District of California signed off after her predecessor, Judge William Alsup, granted preliminary approval last year and has since retired.
The settlement closes a case that had put Anthropic on the hook for potentially far larger statutory damages if it had gone to a jury. Under the Copyright Act, willful infringement can run up to $150,000 per work, and 500,000 works is a number that would have wiped out the company several times over. $1.5 billion, spread across the class, was the price of certainty.
The underlying legal picture is more complicated than the headline number suggests. Alsup split the case in two: he ruled that training an AI model on copyrighted text counts as fair use, a decision widely read as a turning point for the AI industry, but he separated out how Anthropic actually obtained the books. The company had built its training library from two sources — books it purchased and scanned, which Alsup said was fine, and books it downloaded from pirate sites including Library Genesis and Pirate Library Mirror, which he said was not.
Key facts
- 01Judge Araceli Martinez-Olguin signed final approval Monday of Anthropic's $1.5 billion class-action settlement with authors and publishers.
- 02The payout delivers $3,000 per work across an estimated 500,000 works, believed to be the largest settlement in U.S. copyright law history.
- 03Judge William Alsup ruled last year that training on copyrighted text counts as fair use, but that pirating books from Library Genesis was illegal.
- 04Anthropic settled to avoid a jury trial on the piracy question, meaning Alsup's fair-use ruling never reached an appeals court as binding precedent.
- 05Hachette, Cengage, Elsevier, Scott Turow, and S.C.R.I.B.E. filed a similar class action against Google last week over Gemini training data.
That second finding was what pushed Anthropic to settle. Alsup indicated the piracy question could go to trial, and Anthropic agreed to the $1.5 billion figure rather than let a jury set the number. Many of the authors covered by the class do not view the outcome as a win, because the core fair-use ruling went against them and now stands as the most-cited district court decision on AI training data.
Crucially, that fair-use ruling is not binding precedent. It came from a single district court, and by settling before appeal, Anthropic ensured the case will never reach the Ninth Circuit. Other judges hearing parallel cases against Google, Meta, Midjourney, and OpenAI are free to reach different conclusions on their own facts, and several are actively doing so.
The litigation pipeline is deep. Just last week, a group of publishers and authors including Hachette, Cengage, Elsevier, Scott Turow, and S.C.R.I.B.E. filed a class action against Google alleging the company used their copyrighted works to train Gemini. That case, and others like it, will test whether Alsup's fair-use logic holds up when the plaintiffs and the training corpus look different.
The distinction Alsup drew — legal to train on lawfully acquired books, illegal to seed a training library from pirate archives — is now the template every AI lab is being measured against. Anthropic's disclosure that it had pulled from Library Genesis and Pirate Library Mirror was not unusual for the era; those repositories were widely used across the industry during the 2022-2024 training runs that produced today's frontier models. Discovery in the pending cases against Meta, OpenAI, and others is expected to surface similar sourcing.
For Anthropic specifically, the settlement removes a large tail risk at a moment when the company is raising capital at increasingly aggressive valuations and pitching enterprise customers who care about legal indemnification. $1.5 billion is a real number, but it is a known number, and it is spread over a payment schedule the company can absorb. The alternative — an open-ended trial verdict — was the version investors and customers could not price.
The broader industry lesson is narrower than the settlement size implies. Fair use for model training survived, at least in one courtroom. What did not survive was the assumption that provenance of training data does not matter. Every lab that touched Library Genesis, Anna's Archive, or a similar mirror now knows that the acquisition method is a separate cause of action from the training use, and that a jury never had to weigh in for the bill to reach ten figures.
The settlement resolves Anthropic's exposure but leaves the industry's central legal question — is training on copyrighted works fair use, full stop — unsettled and unappealable from this case. The next binding answer will come from whichever of the Google, Meta, Midjourney, or OpenAI cases reaches an appeals court first, and the plaintiffs in those cases have every incentive to press harder than the Anthropic class did. For AI companies still training on scraped or pirated corpora, the Anthropic outcome is not a ceiling on damages. It is a floor.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
