The news
A federal judge in San Francisco granted final approval on July 20, 2026, to the Anthropic copyright settlement, a $1.5 billion deal resolving claims that the AI company downloaded books from pirate libraries, and entered final judgment in the case.
U.S. District Judge Araceli Martínez-Olguín of the Northern District of California signed the order in Bartz v. Anthropic PBC, finding the settlement fair, reasonable and adequate and dismissing the case with prejudice. Judge William Alsup, who oversaw the case earlier and granted preliminary approval, has since retired, TechCrunch reported.
The class covers copyright owners of books on a court-approved Works List drawn from the Library Genesis (LibGen) and Pirate Library Mirror (PiLiMi) collections that Anthropic downloaded. The order puts the list at 482,460 works and cites an expected payout of about $3,000 per work before costs and fees. As of April 16, 2026, claims had been filed for 440,490 of those works, or 91.3%, the order said.
The settlement also requires Anthropic to destroy the original files it downloaded from LibGen and PiLiMi and any copies made from them, subject to legal preservation obligations. According to the order, Anthropic represented that neither dataset, nor any portion of either, was in the training corpus of any of its commercially released large language models.
The release is narrow. Class members give up claims tied to Anthropic's past copying of their books but keep claims over AI outputs and over conduct on or after August 25, 2025, the court said. Earlier in the case, Alsup ruled that training an AI model on copyrighted text counts as fair use, while also finding that Anthropic illegally downloaded and stored millions of books, according to TechCrunch.
Martínez-Olguín also trimmed the fee request. Class counsel Lieff Cabraser Heimann & Bernstein and Susman Godfrey had asked for 12.5% of the fund, or $187.5 million, down from an initial 20% request; the court awarded $101,561,111, nearly 6.8%, and will hold back 10% until a final accounting. It also reduced service awards for the three class representatives to $15,000 each from the $50,000 sought, counted 350 valid opt-outs covering 1,802 works, and overruled all objections.
The numbers
- Settlement fund (non-reversionary)
- $1.5 billion
- Estimated payment per work, before costs and fees
- About $3,000
- Works on the Works List
- 482,460
- Works claimed as of April 16, 2026
- 440,490 (91.3%)
- Attorney's fees awarded
- $101,561,111 (nearly 6.8%)
- Valid opt-outs
- 350, covering 1,802 works
Why CEOs should care
The settlement gives legal and procurement teams a court-approved reference point: about $3,000 per work for books a model developer obtained from pirate libraries. The liability in this case attached to how the books were acquired and kept, not to training itself, which Alsup had found to be fair use, according to TechCrunch. That makes data provenance the question to put to AI vendors. Ask whether any training data came from shadow libraries or other unlicensed sources, what has been purged, and whether contracts carry intellectual property indemnities that cover training-data claims.
For CFOs, the math shows how fast these claims scale: a per-work figure multiplied across hundreds of thousands of books reached $1.5 billion here. A supplier with a thinner balance sheet might struggle to pay a comparable claim, so vendor risk reviews should weigh ability to pay alongside indemnity promises. Boards overseeing in-house model work should ask whether any internal datasets were built from unlicensed copies, since this settlement required destroying such files as well as paying for them.
Claude customers get some comfort, with limits. Anthropic told the court the pirated datasets were not in any commercially released model's training corpus, and the settlement ends the download claims for works on the list. But claims about AI outputs, conduct from August 25, 2025, onward and works not on the list were not released, so legal exposure around generative AI output is still open.
The bigger picture
The court described the deal, quoting its preliminary approval opinion, as the largest copyright class action settlement in history, and said the per-work amount was an order of magnitude above the maximum proposed for books in the rejected Google Books settlement. The settlement does not decide whether fair use applies across the industry, TechCrunch noted, and copyright suits against Alphabet's (GOOGL) Google, Meta Platforms (META), Midjourney and OpenAI are pending. TechCrunch also reported that publishers including Hachette, Cengage and Elsevier, along with author Scott Turow, filed a class action against Google in July 2026 over training for its Gemini models.
What happened next
Between September 2 and 4, 2026, the settlement administrator emailed claim information notices to claimants with valid addresses, according to the settlement website. The site later said a 30-day period for resolving disagreements over how to split payments had been extended to 60 days, and that the parties had told the court the settlement's Effective Date had passed, assuming no late-filed appeal is allowed to proceed.
The administrator expects to send the first round of payments on or before November 15, 2026, for works where claimants agree on the split and have supplied the required information; money for disputed works will be held in escrow. The parties must file a post-distribution accounting within 21 days after the fund is substantially paid out, and class counsel's 10% fee holdback depends on that report.
What “Fact-checked” means
Fact-checking means testing a story’s facts against the evidence before it is published. This story went through at least two separate checks before this version was published.
- What we checked
- Its names, figures, dates, job titles, quotes and who said what were checked against the story’s sources, including its main source where it could be opened. The headline was checked for accuracy and overstatement.
- How
- A first check reviewed the whole story. If it passed, a second, skeptical check went back to the sources to look for mistakes in the most important facts. If a check flagged the story, it was edited to fix the problems found, and a separate re-check then reviewed the whole story again.
- Who
- The checks are made with our newsroom’s technology tools, as steps kept separate from the writing, under rules set by our editor, Hussein Mukhtar. A story the checks still flag is not published automatically; it is held for the editor, who decides whether it is fixed, published or dropped.
- Archive story
- It was written after the event. Its dates were checked against the date of the event, including the dates of any later developments it reports.
- If something is wrong
- “Fact-checked” does not mean error-free. If a material error is found after publication, we correct the story and add a note saying what changed. Report an error
Companies in this story
