Authors challenge publisher and agent claims on Anthropic’s $1.5 billion settlement
Authors are challenging publisher and agent claims on Anthropic’s $1.5 billion copyright settlement, exposing rights-recordkeeping risks for AI litigation.
Authors receiving payments from Anthropic’s $1.5 billion copyright settlement say publishers and literary agencies are claiming money they may not be entitled to collect. The complaints are raising questions about rights records, payment controls and whether the settlement process can reliably distinguish between current publishing agreements and reverted rights.
The dispute comes as payments move forward after a judge approved the settlement in July. The case is an important test of how copyright settlements involving AI model training will be administered at scale—and how authors, publishers and agents will divide compensation when ownership histories are incomplete or contested.
How the Anthropic settlement divides payments
Anthropic agreed to settle a copyright class action after the court drew a distinction between two uses of books. The judge found that training AI models on copyrighted material could qualify as fair use, while obtaining books through piracy was not protected by that doctrine, according to TechCrunch AI’s account of the case.
Under the approved arrangement, authors connected to nearly 500,000 titles are due $3,000 for each qualifying pirated work. A book that remains in print through a traditional publisher is subject to a 50-50 division between the author and publisher.
The rules are different when an author self-published a book or when the publisher’s rights ended because the title went out of print and the rights reverted. In those cases, the author should receive the entire payment. That distinction makes historical rights data central to the process, particularly for older books that have passed through multiple contracts or publishing houses.
The settlement uses August 10, 2022, as the relevant download date. For an author to claim the full amount for a title, the rights reversion must have occurred before that date, TechCrunch reported.
Authors report disputed claims
Several authors said they received notices indicating that another party had filed a claim against money they expected to receive. Mystery and thriller writer April Henry publicly questioned a HarperCollins claim involving a book whose rights had reverted many years earlier. She also said the publisher appeared in a payment-related record as her employer, despite never having been one.
Victoria Strauss, who writes the Writers Beware blog, said the complaints she has received fall mainly into two categories. In some cases, publishers appear to be claiming books for which their rights had already ended. In others, publishers appear to be seeking the full payment even though the settlement would entitle them to only half.
Strauss has cautioned against assuming deliberate misconduct. Some publishers have reportedly described the disputed claims as errors and asked Anthropic to correct them. Authors Guild CEO Mary Rasenberger likewise told The New York Times that she viewed the situation primarily as a consequence of poor recordkeeping and a confusing claims process, rather than an intentional effort by publishers to take authors’ money.
At the same time, Strauss said the volume and repetition of complaints suggested the problem could be broader than ordinary administrative mistakes. That conclusion remains an observer’s assessment, not an independently verified measure of the settlement’s overall error rate.
Claims involving agents add another layer
The controversy is not limited to publishers. Strauss also said she had heard complaints about literary agencies seeking a share of settlement payments. That has surprised some authors because agents generally represent writers in negotiations but are not the copyright owners of the books involved.
Author Courtney Milan criticized the practice in a public Bluesky post, arguing that agents should not claim percentages of the settlement. The available reporting does not establish how many agencies have made such claims, whether those claims are based on individual contracts, or how Anthropic’s claims administrator is evaluating them.
That uncertainty matters for authors and rights teams. A publisher may have a contractual basis to share proceeds from a current title, while an agent’s entitlement could depend on separate commission language. Those questions cannot be resolved solely by identifying the company that once sold or distributed a book; they require reviewing the governing agreement and the date on which rights changed hands.
Evidence, uncertainty and implications for AI litigation
The strongest evidence currently available is a set of author reports, statements from rights advocates and comments attributed to the Authors Guild and publishers. The reporting does not provide a complete audit of claims submitted, the number rejected, or the total value in dispute. As a result, it is too early to determine whether the problem is concentrated among particular publishers or reflects a system-wide failure.
For AI companies, the episode shows that agreeing to a large copyright settlement is only one part of resolving training-data disputes. Payment administration can become a second source of legal and reputational risk when ownership records are fragmented. Future settlements may need clearer data standards, auditable claim histories and a straightforward mechanism for authors to challenge an allocation.
For publishers and enterprise legal teams, the Anthropic settlement highlights the operational cost of rights management. A database that identifies a publisher as associated with a title may not capture later reversions, territorial limits, contract amendments or self-publishing changes. Those gaps can create conflicts even when no participant intended to overclaim.
For authors, the immediate issue is documentation. Writers may need contracts, reversion notices and publication records to show when rights changed. The settlement’s date rule means that a rights history can affect not only who receives payment, but whether the author receives half or the full amount.
What to watch next
The first signal will be whether Anthropic or the settlement administrator publishes clearer instructions on disputed allocations, including how authors should challenge publisher and agency claims. Authors Guild guidance and updates from Writers Beware may also clarify recurring errors and the evidence needed to correct them.
The next important development will be whether publishers identify specific technical or recordkeeping failures and report that disputed claims have been withdrawn or amended. It will also be worth watching for evidence about literary agencies: the number of claims, the contractual rationale offered and whether any are ultimately accepted.
Finally, future copyright cases involving AI model training may use this dispute as a warning about settlement design. Courts and negotiating parties could face pressure to define rights data, claimant verification and appeals procedures before compensation begins, rather than treating those issues as back-office administration.
Creati.ai perspective
The Anthropic settlement is being watched as a copyright event, but its immediate lesson is operational. AI disputes increasingly depend on large datasets and complex ownership histories, and the credibility of a resolution will depend on whether affected people can verify how money was assigned.
The available evidence does not prove that publishers or agents systematically tried to take payments improperly. It does show that repeated rights-recordkeeping errors can produce the same practical result for authors: delayed compensation and a burden of proof shifted onto individuals. For AI builders and enterprise buyers, transparent claims processes and traceable rights data should be treated as part of legal risk management, not as an afterthought.
