Ten and a half months after Anthropic agreed to write the largest check in the history of American copyright litigation, a federal judge has signed off on it. On July 20, 2026, Judge Araceli Martínez-Olguín of the Northern District of California granted final approval to the $1.5 billion class settlement in Bartz v. Anthropic and entered final judgment, clearing the way for payments covering roughly half a million works that authors say were pulled from pirate libraries to build the training corpus behind Claude.
The 23-page order calls the deal "fair, reasonable, and adequate," overrules all 54 objections and comments filed by class members and others, and — in the day's biggest surprise — cuts the attorneys' fee award from the $187.5 million class counsel requested to $101,561,111.
What the Court Approved
Anthropic pays $1.5 billion plus interest into a non-reversionary fund, in installments through September 2027. The per-work recovery is roughly $3,000 before fees, split between authors and publishers per their contracts. The court found the settlement provides "substantial benefits to the Class in light of the novel claims asserted," and noted the figure is "four times the statutory minimum for ordinary infringement" of $750, "which is also the most common award in copyright cases."
The release is deliberately narrow. Class members give up only claims relating to Anthropic's past acquisition and copying of their works — the "inputs" side — through August 25, 2025. Claims based on AI outputs are not released, nor is any claim about future conduct. "A narrow release benefits the Class," the court wrote; members "retain their rights to sue for future misconduct." For authors whose books never made the Works List, claims are "preserved and unaffected by the Settlement" — the distinction from the Google Books settlement rejected in 2011, which released future claims.
Anthropic must also destroy all original files torrented from Library Genesis and Pirate Library Mirror, plus copies originating from them. By the May 14 fairness hearing, claims had been filed on 447,576 of 482,460 eligible works — 92.77 percent. Named plaintiffs Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson saw service awards cut from $50,000 to $15,000 each; the fee award, about 6.8 percent of the fund, reflects a 3.75 lodestar multiplier rather than the 6.92 sought.
Triumph, and Ambivalence
"We are gratified by the Court's ruling granting final approval of this historic settlement. It is the largest known copyright recovery in history," said Justin Nelson of Susman Godfrey. Anthropic read the same record differently: "Training AI on books is fair use under copyright law," deputy general counsel Aparna Sridhar told NPR, adding the company was "looking forward to bringing this matter to a close."
Not every author is celebrating. Graeber, a New York Times bestselling journalist and one of three lead plaintiffs, said he was "proud to be part of a group that showed that a ragtag bunch of authors joined later by publishers could actually hold together as a class, face a Goliath like Anthropic and get a meaningful number out of them." But he called the past two-plus years costly: "A lot of travel, a lot of discussion about what to do and how to proceed and a lot of jobs passed up. I'm much poorer for this settlement, ironically." He expects roughly $3,100 per book.
Objectors were pointed. Robert Jacobson argued $3,000 is a fraction of the Copyright Act's $150,000 ceiling; Victoria Pinder, author of 40-plus novels registered under one group copyright, objected that she stood to receive a single share rather than 40. The Authors Guild itself called $3,000 "paltry in light of the gross theft." The court overruled each, observing that "a multi-billion dollar verdict could create its own major risk for the class"; against that risk, the settlement "provides value in prompt closure."
What It Settles, and What It Doesn't
Industrywide, very little — the uncomfortable truth beneath the record number.
The architecture traces to Judge William Alsup's June 2025 ruling, which found Anthropic may have downloaded as many as seven million books from pirate sites. Alsup held that training a large language model on copyrighted text was "exceedingly transformative" and therefore fair use, as was digitizing lawfully purchased print copies. What he refused to bless was the sourcing: building the library from LibGen and PiLiMi was not fair use, and that question was headed to trial. Anthropic settled rather than let a jury price seven million infringements at up to $150,000 apiece.
Because it settled, the case never reaches the Ninth Circuit. Alsup's training-is-fair-use holding remains one district judge's view — persuasive, not binding — and he has since retired. Meta prevailed in a parallel authors' case last year, but on a narrow evidentiary failure: plaintiffs did not show market harm, not that none exists. Suits against OpenAI, Stability AI, Midjourney, and Google continue; in July, a coalition including Hachette, Elsevier, and author Scott Turow filed a fresh class action against Google over Gemini.
What Bartz establishes is a price tag on one behavior: torrenting training data at industrial scale costs roughly $3,000 a book plus destruction of the files. It says nothing definitive about licensed acquisition, outputs, or the market-dilution theory authors increasingly press. Bartz put that open question plainly to NPR: "The algorithm is being used to essentially try to put us out of a job."
What to Watch
Three things. First, distribution timing: no payout date has been announced, and Special Master Theodore K. Cheng will resolve author-publisher disputes under seal. Class counsel committed to taking fees in step with money actually reaching claimants, and a post-distribution accounting — for which ten percent of the fee is withheld — could shave it further. Second, whether any objector appeals; final judgment starts that clock, and an appeal delays checks. Third, whether the next court to rule on AI training reaches for Alsup's transformativeness framing or for the market-harm analysis authors have been sharpening since. Nothing in the July 20 order requires either.
Legal conclusions described here are the court's or named counsel's; this is reporting, not legal advice.
“A lot of travel, a lot of discussion about what to do and how to proceed and a lot of jobs passed up. I'm much poorer for this settlement, ironically.”— Charles Graeber, Named plaintiff and bestselling author