A federal judge has given final approval to the Anthropic copyright settlement, a $1.5 billion class action payout to authors and publishers, while simultaneously trimming the lawyers’ payday by roughly $86 million. The cheques can now be cut. Whether anyone feels vindicated is a different question.
Judge Araceli Martinez-Olguin of the U.S. District Court for the Northern District of California signed off on the settlement after Judge William Alsup, who had handled most of the case, retired. Alsup had previously ruled that Anthropic illegally downloaded and stored millions of copyrighted books from pirate sites including Library Genesis and Pirate Library Mirror.
What the Anthropic Copyright Settlement Actually Pays Out
Across an estimated 500,000 works, rights holders will receive approximately $3,000 per work, shared between authors and publishers. Martinez-Olguin noted in her ruling that this figure is ‘four times the minimum statutory damages amount for willful infringement’ and four times the statutory minimum for ordinary infringement, the latter being $750, according to Publishers Weekly. So by one measure, authors did reasonably well. By another measure, the core legal question went against them.
Alsup had ruled that training an AI model on copyrighted text counts as fair use, describing Anthropic’s use of the books as ‘exceedingly transformative,’ per Deadline’s reporting on that ruling. That framing was widely seen as a significant moment for the AI industry. It was also the part authors objected to most loudly, because the settlement preserves Alsup’s fair use reasoning while letting Anthropic pay its way out of the piracy question.
The piracy question was the sticking point. Anthropic built its training library from two sources: books it purchased and scanned, and books it downloaded from pirate sites. Alsup found the second method illegal and left open the prospect of a jury trial on damages. Had the case proceeded that far, Anadolu Agency reported that Anthropic faced potential exposure of up to $150,000 for each willfully infringed work. Against that backdrop, settling starts to look less like generosity and more like prudent risk management.
Plaintiffs’ attorney Justin Nelson called the result ‘the largest publicly reported copyright recovery in history.’ Plaintiffs’ firm Lieff Cabraser noted the settlement received ‘overwhelming support from writers and publishers’ after a year of litigation covering discovery, class certification, and summary judgment. Martinez-Olguin herself reasoned that ‘success at trial was not assured, and a loss would have left the Class with no recourse.’
The suit was originally filed in 2024 by authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson, who alleged Anthropic used unauthorised copies of books while developing its Claude AI models.
One other adjustment Martinez-Olguin made: class counsel had requested $187.5 million in fees. Law.com reported the judge cut that to $101.6 million, a reduction of approximately $86 million. Not everyone left the room happy.
Why This Settlement Changes Very Little for the Rest of the Industry
Because Anthropic chose to settle, the case will never reach an appeals court. Alsup’s fair use ruling remains a single district court decision, not binding precedent. Every other judge facing an AI copyright case can reach their own conclusions on their own facts, and several are doing exactly that.
Google, Meta, Midjourney, and OpenAI all face active copyright litigation over AI training. The most recent addition arrived this month, when Hachette Book Group, Cengage Learning, Elsevier, author Scott Turow, and S.C.R.I.B.E. Inc. filed a class action complaint against Google in federal court. The filing date carries a minor discrepancy: the International Publishers Association records it as 10 July 2026, while Hachette’s own newsroom gives 13 July 2026, possibly reflecting the gap between initial filing and public announcement.
The complaint, filed via the Association of American Publishers, alleges Google illegally copied books and journal articles, downloaded unauthorised web scrapes including material from known pirate sources and behind paywalls, and deliberately stripped copyright management information from the works to conceal the fact that its Gemini models were trained on, in the complaint’s framing, stolen materials. That last allegation goes further than anything in the Anthropic case, and it will be considerably harder to wave away with a fair use argument.
The Anthropic settlement closes one chapter neatly. The broader legal contest over what AI companies can legally feed their models is only getting louder, and the next ruling that matters will come from a court that cannot simply be settled out of.
