Books are not like music downloads, and authors are not like recording artists, but the class-action settlement between Anthropic and book authors has produced the kind of accounting fight that the music industry has been through before. At stake is $1.5bn. The money is compensation for hundreds of thousands of books that Anthropic was found to have pirated, copied into its systems and used to train Claude, the company's flagship AI assistant.
Under the settlement, which a federal judge in the Northern District of California approved in July, each covered title can generate as much as $3,000. The total list reaches roughly $1.45bn at the top rate. Authors and publishers began filing claims expecting a straightforward payout. What they got instead was a tangle of competing claims over the same titles, with a court-appointed administrator telling both sides that they disagree.
The default split between authors and publishers
No author keeps the full $3,000. The settlement's default split is 50/50 between the author and the publisher who holds the rights, and co-authors may further divide the author's share. That split is not the result of individual contracts; the class-action council set it after taking input from the Authors Guild and from major publishing houses. Nothing about it was negotiated title by title, which means the terms of one book contract can look completely different from another when the one-time payment lands.
Textbook authors are the ones complaining most loudly. Some contracts give textbook authors only 10 to 15 percent of the total fee. Mary Rasenberger, the chief executive of the Authors Guild, said those royalty structures were designed for a different kind of revenue stream, not for a one-time legal settlement over an AI company's unlawful copying. Rasenberger said textbook authors are “not happy right now.” She is not accusing publishers of a coordinated land grab. Her real fear is that publishers are simply claiming titles they should no longer have on their lists, including works that reverted to authors years ago but never left the publishers' internal catalogues.
One author who had to prove she owned her own book
April Henry is an author of more than 30 mystery and thriller novels. When she logged into the settlement claim portal, she found that HarperCollins was listed as a part-owner of her first book, “Circles of Confusion,” published in 1999. In 2007, the rights reverted to Henry. She said on Threads that she had to produce proof. Her agent sent a letter confirming she was the sole rights holder, and she uploaded it to the portal. Only then did the portal show the entire $3,000 going to Henry. She does not believe HarperCollins was deliberately trying to deceive her, but the incident is an emblem of the settlement's wider problem. HarperCollins declined to comment.
Henry has 22 titles on the list of books Anthropic used to train Claude. After publishers and co-authors take their shares, she expects to receive somewhere in the mid-$20,000 range. Payments were expected to begin in August, she said, but they have not arrived.
Another author whose publisher agrees on ownership, but not the split
Amy Lupold Bair found herself in a different and arguably more frustrating position. Her publisher does not dispute that she owns the rights to her guidebooks about blogging and family life online. What the publisher disputes is the percentage she should receive. As Bair put it on Threads, the publisher wants the author, whose entire work was taken, to receive only 10 percent. She did not name the publisher in the post, but her website lists John Wiley & Sons as the publisher for two of her titles, “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. Wiley did not say whether the Dummies series counts as educational. The Association of American Publishers declined to comment.
The date that decides who gets paid
At the center of both disputes is a question about time, not authorship. The relevant owner is not the person who holds rights today. It is the person who held the rights when Anthropic downloaded the book. Rasenberger dates the infringement to 2021 and 2022, when Anthropic was assembling its training data. A title that reverted to its author last year is still the publisher's if the publisher owned it at the moment the book was copied.
The Authors Guild said it heard from an author this week who believed a recent reversion would solve everything. The rights had come back to the author this year, but the download had happened four years earlier. The reversion settled nothing. The publisher of record at the time of infringement is the one who receives payment.
That reality turns what should be a compensation event into a records exercise stretching across decades of book contracts. New authors sign with modern boilerplate that may or may not mention digital rights. Older authors signed contracts that say nothing about AI training, text data mining, or copyright settlements. Where an author and a publisher cannot agree on the ownership history or the correct contractual allocation, the dispute goes to a court-appointed arbitrator. The Authors Guild says it will fight for any author it believes is losing money it is owed.
Publishing's streaming moment
Kristelia García, who teaches copyright at Georgetown Law, compares the Anthropic settlement with the case brought around Eminem's digital royalties. Producers sued a Universal Music Group subsidiary in 2007, claiming that downloads should produce a higher royalty rate than physical records, and the case settled in 2012. The parallel, García says, is in the paperwork. Most book contracts do not mention copyright settlements. They do not mention revenue generated by technologies nobody had dreamed of when the author signed. Publishing is having a streaming-like moment in which its old contracts do not anticipate the current dispute, she said.
What Anthropic still owes and what it says
Anthropic agreed to settle only after a judge allowed the case to proceed to trial. That judge found that stockpiling pirated books gave the authors a valid claim. The same ruling also held that training Claude on books the company had legally purchased was fair use. That second finding has become the cornerstone of nearly every AI defendant's argument today, which is why technology firms have started buying up old book catalogues and out-of-print titles.
Anthropic has not publicly commented beyond an earlier statement. In May, its deputy general counsel, Aparna Sridhar, said more than 91 percent of eligible authors and publishers had already claimed their share. Anthropic wanted the matter closed. The latest ownership disputes surfaced after that statement, and the settlement administrator is now working through them.
Anthropic still faces other copyright troubles. Sony Music and Warner Chappell are suing over song lyrics in Claude's training data. In New York, Microsoft has argued that its Copilot tool almost never reproduces entire books, deploying a fair use defense similar to the one that helped Anthropic. News publishers and the Authors Guild have pushed back against that argument.
What is left, then, is more complicated than Anthropic had hoped. The settlement was meant to be the case that demonstrated how AI companies could compensate authors without years of individualized litigation. Instead, it has produced a list of 482,000 books, an arbitrator, and thousands of authors digging through contracts they signed before Claude existed. The money is real, but the allocation is now its own slow and messy chapter.
Source:TNW | Legal News