Anthropic has agreed to pay as much as $1.5 billion to settle a copyright case brought by authors who say their books were quietly taken and used to train the company’s AI assistant Claude. The payments are now being calculated, and the process has exposed an uncomfortable reality: a substantial slice of the money intended for authors may end up with their publishers.
The settlement covers more than 482,000 books. It offers up to $3,000 for each title that a judge found Anthropic downloaded illegally and stored while building Claude. The class action was filed in 2024, and a judge in the Northern District of California approved the terms in July. At the top rate, the full list of titles would be worth roughly $1.45 billion.
The announced payout figure never meant that each author would receive a direct cheque for $3,000. Publishing contracts, co-authorship arrangements, and reversion clauses determine who qualifies. Publishers have filed claims on the same books their authors are claiming, and the settlement administrator has begun telling both sides they disagree. Reports on Friday described the standoff as one of the most complicated parts of the process.
How the settlement is divided
Authors do not keep the entire settlement award. They generally split it with publishers to whom they granted rights, and with co-authors where a book was written by more than one person. The percentages were not negotiated for each title. A committee connected to the class action set them after consulting the Authors Guild, which says it has more than 18,000 members, and several publishing houses. Nobody went through each contract to reconcile the percentage splits with the ownership listed in the settlement file.
Textbook and educational authors are likely to receive the smallest slice. Mary Rasenberger, chief executive of the Authors Guild, says those writers often have contracts that grant them only 10 to 15 percent of the amount paid for their work. “It’s the textbooks where there are a lot of unhappy authors right now,” Rasenberger said. She nonetheless described the pattern as something other than a publisher land grab. “I don’t see this as a grab by the publishers.” Her actual worry is record keeping: many publishers never removed reverted titles from their catalogues and are now claiming them by default.
The distinction matters because a copyright settlement is not a normal royalty payment. It is compensation for an unauthorised use of an entire work. Authors who kept only a small share of their book’s proceeds may feel that the windfall should belong to them in full. But the rights they signed away decades ago often gave publishers broad authority over licensing and infringement claims. That authority did not disappear simply because the book stopped selling.
Two cautionary stories from the claims portal
The practical consequences are visible in the experience of April Henry, a novelist with more than thirty mysteries and thrillers. When she logged into the claims portal, she saw HarperCollins listed as a part-owner of her first book, Circles of Confusion, published in 1999. Henry’s rights to the book reverted to her in 2007. After she flagged the issue on social media, her agent produced a letter confirming that she owned the rights. She uploaded the letter, and the portal eventually showed her receiving the full $3,000 award for that title. “I don’t think Harper was deliberately trying to cheat,” Henry said. HarperCollins declined to comment.
Henry has twenty-two titles on the list of books Anthropic allegedly took. Even with the corrected reversion record, she expects her total payout to land somewhere in the mid-$20,000 range after her publishers and co-authors take their shares. She was originally told that payments might start arriving in August, but they have not materialised yet. The delay is partly because the administrator is still being asked to adjudicate competing claims of this sort.
Amy Lupold Bair, who writes guidebooks about blogging and family life online, is facing a different issue. Her publisher is not challenging her ownership of the books. It is challenging the split. Bair wrote on social media that the publisher “only want me, the author whose entire work was stolen, to get 10%.” Her website lists John Wiley & Sons as the publisher of Raising Digital Families for Dummies and Blogging for Dummies. Wiley said it had filed claims for all Wiley published works and that allocations for educational titles follow individual contracts. The company did not say whether it classifies the Dummies series as educational. The Association of American Publishers declined to comment.
The date of downloading, not the date of payment, may decide ownership
Underneath the competing claims is one legal question: who owned each book when Anthropic downloaded it, not who owns it today. Rasenberger places most of those downloads in 2021 and 2022. That means a title that reverted to its author in the past year could still belong to the publisher for the purpose of this case because the publisher held rights when the infringement occurred.
The Authors Guild heard from an author this week who received their rights back only this year. That fact settles nothing, because Anthropic’s download predates the reversion by four years. The entire exercise has therefore turned into something closer to an audit of decades-old publishing contracts. Where the two sides cannot agree, a court-appointed arbitrator will decide. The guild says it plans to fight hard for any author it believes is being treated unfairly.
Contract language is not the only problem. Many authors have changed publishers, died, or been acquired by other companies. Some catalogues were bought and sold without a clear record of what rights travelled with them. Reversion letters may have been sent but never recorded in the right database. Authors who self-published a later edition may not know whether the earlier contract was still active. These recordkeeping gaps are now affecting the distribution of an AI settlement worth billions.
A lesson borrowed from the music business
Copyright lawyer Kristelia García, who teaches at Georgetown Law, compares the Anthropic settlement to the dispute over Eminem’s digital royalties. In 2007, producers sued a Universal Music Group subsidiary, arguing that downloads should produce a higher royalty rate than physical records. That case settled in 2012, but it became a landmark example of contracts failing to anticipate future distribution technologies.
Book contracts, García notes, say almost nothing about copyright settlements. They certainly say nothing about revenue from AI training or from technologies that did not exist when the authors signed. Publishing is now having the kind of “streaming moment” that music had years earlier, when recorded contracts were ill-equipped to handle digital licensing. The similarity is not in the amount of money at stake but in the paperwork required to resolve it.
What comes after the settlement
Anthropic agreed to settle only after a judge permitted the case to go to trial. Crucial findings were already in place. The court ruled that stockpiling pirated books was enough to give authors a claim against the company. The same ruling also found that training Claude on books Anthropic had legally bought was fair use. That second finding has become a touchstone for nearly every AI company facing copyright litigation. It is one reason firms have recently begun buying old printed books and building digital corpora from them.
Anthropic has pointed to a statement it issued in May, when its deputy general counsel Aparna Sridhar said that more than 91% of authors and publishers had already claimed their shares. The company said it wanted the matter closed. The recent ownership disputes surfaced after those claims were filed, making the settlement more difficult to complete than Anthropic may have expected.
Anthropic also has other copyright problems. Sony Music and Warner Chappell have sued over song lyrics allegedly included in Claude’s training data. In New York, Microsoft has argued that its Copilot system almost never reproduces books, using a fair use defence similar to Anthropic’s. That case is being pursued by news publishers and the Authors Guild. Each new dispute will turn on the same kinds of records and contract terms now being examined in Anthropic’s settlement.
The Anthropic settlement was designed as a clean way to show that AI developers could compensate rights holders. Instead, it has produced a long list of 482,000 books, a new arbitration process, and many authors rereading contracts they signed decades before Claude existed.
Source: TNW | Legal News