By This Hour AI Desk
Authors expecting compensation from Anthropic’s $1.5 billion copyright settlement are challenging claims made by publishers and literary agencies against payments tied to their books. The disputes center on a deceptively difficult question: who held the relevant rights when the works were downloaded.
For writers, the answer can determine whether they receive the full payment for a title, half of it, or must first contest another party’s assertion. Reports of claims involving books whose rights had allegedly returned to authors have prompted concern that the settlement’s distribution process may be misreading old publishing arrangements. The allegations have also exposed a tension between the scale of the payout process and the uneven state of rights records that can span decades.
The available accounts do not establish that publishers or agents deliberately sought money to which they were not entitled. Some publishers reportedly acknowledged errors and asked Anthropic to amend them. The Authors Guild’s Mary Rasenberger has characterized the problems as a consequence of poor recordkeeping and a complicated process rather than an effort to deprive writers of payment. But Victoria Strauss of Writers Beware said the recurrence of similar reported errors could point beyond ordinary isolated mistakes.
Payment rules turn on a book’s rights status
The settlement concerns pirated books used in connection with Anthropic’s AI training, as described in the supplied reporting. It followed a copyright class action in which a judge found training AI models on copyrighted material to be permissible under fair use, while finding the acquisition of pirated material impermissible. The settlement received final approval in July, allowing payments to proceed.
Nearly 500,000 titles are described as eligible for $3,000 for each pirated work. That headline figure does not mean every eligible writer receives the same amount. Distribution depends on the title’s publishing and rights status, making ownership history central to the claims process rather than a secondary administrative detail.
Where a traditionally published book remains in print, the payment is described as split equally between its author and publisher. A self-published title is treated differently: its author is described as entitled to the entire payment. So is a book whose publisher allowed it to go out of print and returned the rights to the author, provided the relevant reversion occurred before the settlement’s specified date.
That distinction creates a sharp dividing line. A publisher with continuing rights may be entitled to half the compensation under the reported terms. A publisher whose rights had already reverted would not have the same claim. The difference is particularly consequential because a claim by another party can force an author to challenge the allocation before payment is resolved.
The source material identifies August 10, 2022, as the settlement’s download date. For an author to assert a 100% claim on a reverted book, the rights must have reverted before that date. A reversion after it would therefore not necessarily produce the same result, even if the author controls the book today. That timing rule helps explain why disputes may require examination of specific contracts, notices, and the history of an individual title rather than broad assumptions about a writer’s relationship with a publisher.
Authors describe claims on reverted titles and full payouts
Writers who received notices that another party had made a claim have taken their concerns public. The complaints described in the reporting generally fall into two categories. In one, a publisher allegedly sought a payment for a book after rights had reverted to the author. In the other, a publisher allegedly claimed the entire $3,000 payment when the reported settlement framework would provide only a 50% share for an in-print traditionally published work.
April Henry, a mystery and thriller writer, publicly alleged that HarperCollins claimed payment connected to one of her books even though the rights had reverted many years earlier. That is an allegation by Henry presented in the reporting, not a finding that the claim was improper. Still, it illustrates the practical difficulty confronting writers: an apparent mismatch can arise long after a contract was signed and long after a work’s commercial status changed.
Strauss said complaints reaching Writers Beware broadly tracked those two patterns: claims involving reverted works and claims seeking a full payment rather than half. The reports alone cannot show how representative the incidents are across the nearly 500,000 eligible titles. They do, however, describe repeat problems at the point where historical publishing data meets a large-scale settlement administration system.
The distinction matters because rights reversion is not simply a label attached permanently to a book. A title may have been published traditionally, gone out of print, and later been republished or otherwise managed under a changed arrangement. Payment eligibility in this settlement, as reported, rests on the rights position before the designated August 2022 date. Sorting that record could be straightforward for some books and difficult for others.
Recordkeeping explanation faces questions over recurring errors
Rasenberger, chief executive of the Authors Guild, said she did not see the situation as an intentional attempt by publishers to take money from authors. Her explanation was more administrative: poor records combined with a confusing claims procedure. That account is consistent with reports that some publishers have recognized mistakes and asked Anthropic to fix them.
It also speaks to the sheer complexity of processing claims connected to a large backlist of books. Publishing rights can change hands, lapse, or revert. The available information does not say what records publishers submitted, how claims were screened, or how often competing submissions arise. Without that information, it is not possible to determine whether the reported cases originated with publisher databases, settlement forms, administrative review, or some combination of those factors.
Strauss has left room for the recordkeeping explanation while warning against dismissing the pattern too quickly. She said poor records could plausibly explain errors, but argued that the number of similar reports over a short period suggested a potentially widespread or systemic issue rather than routine glitches. These views are not necessarily incompatible. A systemic problem can stem from systemic data failures rather than intentional conduct. But the distinction matters to authors seeking a remedy and to any assessment of responsibility.
No figure in the supplied reporting establishes how many payment claims were wrong, how many were corrected, or how many reflected a genuine disagreement over rights. Nor does it establish whether any disputed claim was knowingly improper. The reported examples show that authors have identified apparent conflicts; they do not, by themselves, measure the full scale or cause of the problem.
Agency claims add a separate dispute over entitlement
The complaints are not confined to publishers. Strauss said Writers Beware had also received reports of literary agencies making claims on settlement payments. The source material says agents are not rights holders in the books they sell, making such claims a separate source of concern for authors.
That description does not resolve what authority, if any, an agency may have had under an individual representation agreement or in submitting paperwork on a client’s behalf. The supplied material does not provide the terms of any agency contract, identify how many agencies made claims, or establish the basis on which a particular agency acted. What it does show is that authors and advocates see a difference between representing a writer in a deal and holding the underlying rights to a book.
For writers, the immediate route described in the reporting is to dispute an allocation. That process is especially important for an author who believes a reverted work qualifies for the full payment. A dispute can place the rights history before the settlement mechanism rather than leaving an apparent allocation unanswered.
The broader consequence is that settlement administration has become another arena in the debate over how AI companies use copyrighted works. The underlying case dealt with the legality of different stages of acquiring and using books for AI training. The current complaints concern a later question: whether the people identified to receive compensation accurately reflect the contractual ownership of each work. Other disputes over AI training have focused on alleged use of copyrighted journalism, including a separate case involving claims by Seattle Times and Newsday against OpenAI and Microsoft. The Anthropic matter shows that even after a settlement is approved, questions over rights and payment can persist.
Authors, publishers, agencies and the settlement process now face a practical test. Each challenged title may require the parties to establish whether rights were active, reverted, self-published, or otherwise allocated on the relevant date. Clear corrections where records are wrong could resolve individual cases, but the reported repetition of complaints raises a larger question about whether the process can reliably identify ownership across a vast catalogue.
The report underlying these accounts has not been independently corroborated. The available material supports that authors have raised these allegations and that competing explanations have been offered; it does not establish the number of erroneous claims, the intentions behind them, or the ultimate resolution of any individual dispute.
Reporting notes
What is confirmed: Some publishers reportedly acknowledged mistakes, while authors and advocates dispute whether recurring errors are isolated or systemic.
Why this matters: A title’s rights status can determine whether an author receives all or half of the reported $3,000 payment.
What remains unclear: The scale, cause and intent behind disputed claims have not been established. This report is based on one source and has not been independently corroborated.
