Authors owed $3,000 for each of their pirated books under Anthropic's copyright settlement are discovering that someone else has claimed the money. April Henry, a mystery and thriller writer, said HarperCollins claimed the payout on a book whose rights reverted to her at least 17 years ago; the same day, she received a notification that HarperCollins had been added as her employer, which it never was. Victoria Strauss, who writes the Writers Beware blog, says she is now receiving two distinct kinds of complaint, and that literary agencies — which hold no rights in the books they sell — are also claiming a share.
The terms are simple enough to state. Anthropic settled the class action last year, after a judge ruled that training AI models on copyrighted material is permitted under fair use, but that acquiring that material through piracy is not. Final approval came in July, which allowed payouts to move to the next stage. Authors of nearly 500,000 works receive $3,000 per pirated title. Where a book is still published by a traditional publisher, the payment is split 50/50 between author and publisher. Where the author self-published, or the publisher returned the rights and let the book go out of print, the entire sum belongs to the author. Nearly 500,000 works at $3,000 each puts the fund at roughly $1.5bn.
Strauss's two categories map directly onto the two ways that arrangement can go wrong. Publishers are demanding payment for works they no longer hold any legal rights to, because those rights reverted to the author. And publishers are demanding the full 100% of a payment when they are entitled to half.
She has been careful about motive, saying she does not want to explain by malice what can plausibly be explained by poor record-keeping, and noting that some publishers have already acknowledged errors and asked Anthropic to correct them. Mary Rasenberger, head of the Authors Guild, told The New York Times that she does not see this as publishers trying to seize the payouts and does not think they are deliberately setting out to harm authors; in her view the situation was a predictable product of inaccurate records and a confusing distribution process.
Strauss is less willing to leave it there. She accepts that the complaints she has seen are a small fragment of what is happening. But the unusually large number of reports over the last two days, and the repetition of the same errors, point in her assessment not to the ordinary failures you would expect from an operation this size but to something more widespread and systemic.
The agency claims are stranger still. An agent is not a rightsholder in the book they sold. Courtney Milan — the pen name of Heidi Bond, a former judicial clerk and law lecturer — put it more bluntly on Bluesky, writing that some agents appear to be trying to take a percentage of the Anthropic payouts when they should not be.
The charitable reading and the damning one are not actually in conflict, and this is where the settlement's design starts to look like the problem. Nobody has to be acting in bad faith for the outcome to be systematically tilted. A flat $3,000 per work is administratively elegant and made the settlement possible at this scale; it also meant the hard question — who owns what, right now, across half a million titles and decades of contracts — was never answered by the court. It was pushed downstream to the claims process, where the party with the institutional capacity to file at volume is the publisher, and the party who has to notice, object and prove is the author. Bad records produce errors in every direction in theory. In practice they produce errors in the direction of whoever fills in the form.
Milan and the Authors Guild have both explained how authors can dispute their allocation, and the disputes turn on a single question: when the rights came back. To claim the full amount, an author has to establish that reversion happened before 10 August 2022, the date the settlement calls the "download date".
That date does an enormous amount of work. An author in April Henry's position, with rights back for at least 17 years, has to document a reversion that far back or forfeit half of $3,000 — while the publisher whose records were bad enough to produce the claim in the first place keeps it. Nothing in the process as described imposes any cost on a claim that turns out to be wrong. A settlement built to punish one company for taking books it had no right to has arrived at a claims process where the burden of proving ownership falls on the people who wrote them.