Executive Summary

Key Numbers

Sources: TechCrunch (2026-09-06), Writer Beware (2026-09-04), Authors Guild (2026-09-04)

The case started with a class action three authors brought against Anthropic. Judge Alsup held that training an AI model on copyrighted writing is itself fair use, and that downloading those books from pirate sites and keeping them is a separate matter. Anthropic chose a $1.5 billion settlement over a jury trial. Alsup retired in the interval, so final approval was signed by Judge Martinez-Olguin on July 20.

Approval turned the case into an accounting job. The settlement administrator sent notices to all claimants this week. Each notice lists the works that claimant filed for, along with who else filed a claim on the same work and what percentage they asked for. Most claimants only have to confirm what is there. Where co-claimants entered percentages that conflict, the notice routes them to a portal to revise the numbers or confirm them.

Four paths decide the percentage. A book that was self-published, or whose contract terminated, or whose rights reverted, pays the author in full. A book still in print under a live contract splits evenly between author and publisher. That default comes from common publishing contract language that divides any recovery in copyright infringement litigation equally between the two. The fourth path, educational works, sits outside the default altogether, and that exemption returns in section 3.

The rule turns on a single fact. Who held the rights to this book on August 10, 2022? The Authors Guild's position is that a reversion before that date entitles the author to 100 percent of the funds for the title, and that a reversion after it may leave the publisher with a valid claim, since the publisher was the legal owner when the infringement occurred, which drops the author to 50 percent. One line of fact, and the document that would fix it is usually scattered in an author's drawer or a publisher's filing cabinet.

Examples poured onto social media as soon as the notices went out. Mystery and thriller author April Henry posted on Threads: "WTF is HarperCollins playing at? They claimed one of my books on the Anthropic Settlement that reverted back at least 17 years ago."

Victoria Strauss, who has tracked publishing scams and industry trouble for years at the blog Writer Beware, sorted the reports reaching her into two categories: publishers making 50 percent or 100 percent claims on rights-reverted works, where they should receive no share, and publishers making 100 percent claims on in-print works, where their share should be 50 percent. Some of it was too large to read as individual slips. One author told her that 16 of their rights-reverted books were being claimed. Another had 11.

Within a day Strauss had an alphabetical tally running to 25 publishers in the first category and 22 in the second, not counting the imprints in parentheses, and she notes that she is adding to the lists as information comes her way. The rights-reverted list includes HarperCollins, Penguin Random House, Simon & Schuster, Hachette and Macmillan. The in-print list holds six university presses, among them Columbia, Georgetown, Johns Hopkins and Rutgers, which Strauss flagged in an aside: "Odd how many university presses are in this category." She does not read that second category as error across the board. For 100 percent claims on in-print books, "in some cases that may reflect the publisher's ownership of copyright. Where the author holds copyright, it's harder to figure what's going on."

Strauss herself does not put the cause down to malice. She wrote that she is "always reluctant to attribute to malice what can be plausibly explained by poor recordkeeping (or poor checking of that recordkeeping)," and she suspects poor records, poor checking, or overworked or inexperienced staff assigned to the task. She attached a condition to that: none of it is an excuse, especially where multiple books are involved.

Author Courtney Milan, who had been watching the same scene since before the Writer Beware post went up, read the situation more sharply. On Bluesky, Milan wrote that publishers appear to be "[n]ot doing a good job (or possibly any job) of checking to see if rights have reverted on a work before claiming them in the settlement." Milan had just found a claim on one of their own books, a title whose rights reverted in late 2020, and had sorted it out by uploading a copy of the reversion letter. Milan's guess at why skipping the check gets chosen: "this is the kind of laziness that gets chosen because then they (a) don't have to do any work and (b) if someone doesn't argue then oops! they get $1500 more, well, that's a shame." That $1,500 is half of the $3,000, the share that moves to the publisher instead of the author.

Several publishers did call it a mistake. Strauss reached out to Kensington Publishing CEO Steve Zacharius after several of his authors reported 100 percent claims; he indicated that it wasn't Kensington's intent to claim 100 percent, and told her that Anthropic was aware of the issue and was correcting it. Three other publishers, contacted by their authors, said essentially the same thing, and McFarland said the same on Bluesky. The Authors Guild notice carries the same account. Certain publishers have informed the settlement administrator that they mistakenly selected a 100 percent allocation rather than the default option, the administrator is updating those percentages, and authors who suspect their publisher made that selection should log in again next week.

With the explanations stacking up, Strauss raised another possibility. Given how many authors were reporting the exact same wrongful claim from many different publishers, she wondered whether, rather than greed or evil, some kind of glitch in the settlement claims system was responsible, or whether it was an artifact of batch claims by publishers. She said she did not know. If that guess holds, the cause sits deeper than a person picking the wrong option one book at a time. It would mean a list went in whole, and the list did not carry the current state of the rights.

For authors who landed on the No Disagreements page, Strauss's advice was to expand every title and read the claimants anyway: "Don't assume, just because the page says No Disagreements, that everything is all good! Upon checking, you may discover claims you didn't know about with which you definitely disagree. (I did.)" The Authors Guild is more specific. Claimants who receive a notice stating that "All claimants currently agree on the allocations" when that is not correct, for instance an author who filed for 100 percent and now sees an agreed 50-50 default, should immediately contact the settlement administrator and say that a disagreement does exist. The reconciliation pass built to surface conflicts was reporting no conflict.

Mary Rasenberger, CEO of the Authors Guild, told The New York Times that she doesn't see this as "a grab by the publishers" and that she doesn't believe publishers are "specifically trying to screw any author over." She argued instead that this is the predictable result of bad recordkeeping and a confusing settlement process. The organization that watches the industry and the organization that speaks for authors land in the same place on cause.

Educational publishing shows most sharply what happens when the record is absent. The settlement exempts from the default split any publisher that specializes in creating and publishing works, including but not limited to textbooks, for the instruction of students and professionals, distributed for and through educational and professional markets. Claimants on that side had no default percentage to select in the first place, so authors and publishers entering different numbers was far more common.

The Authors Guild explains the background through the shape of the contracts. A standard trade agreement typically licenses specific uses exclusively and expressly reserves the rest of the rights to the author. Educational agreements are often all rights assignments, where the publisher assumes full ownership of the copyright and the author retains nothing unless the contract says so. On top of that, most trade contracts carry an express provision on infringement actions and how an award is allocated, and very few educational contracts do. Several educational publishers filled that blank with the royalty rate in the contract. If the author's royalty is 10 percent, the publisher claims 90 percent of the award. Writer Beware heard from authors with Wiley and Pearson that the publishers were claiming between 75 and 90 percent, and Strauss noted that this isn't necessarily a wrongful claim if the contract reads that way.

The same blank got filled at literary agencies. The class in this settlement is limited to rightsholders, and an agent is not a rightsholder in the books they sell. Even so, in a single morning Writer Beware heard from five authors with six different agencies that had claimed between 15 and 25 percent of the allocation on books they repped. The Authors Guild confirmed that some agents have filed claims for 15 percent and stated it plainly: agents are not right owners and as such are not legal claimants entitled to payment from the claims administrator. An agent who believes a commission is owed has to obtain it from the author.

Not all of those claims were intended, either. One agent, whose authors found the agency listed as a claimant, told Strauss they had not signed up for anything, and that on many of their authors' statements the publisher lists the agency as the recipient, "probably due to publishers pulling the payee info from the agency clause in the agreements which generally does send all income to the agent on the author's behalf." That clause designates a payment route on the author's behalf. It records where the money goes, not who holds the rights, and at the claim stage the two got read as the same thing.

Strauss's separate list of other discrepancies reads as a catalog of works whose identity and rights nobody ever wrote down in a single line. A book originally published in the UK, where the author and the UK publisher each properly claimed 50 percent, drew a 100 percent claim from the US publisher. A rights-reverted book with an outstanding audio contract drew a 100 percent claim from the audio publisher. Works from a defunct publisher were claimed by a company the authors had never heard of. Claims around anthologies collided with claims on the individual stories inside them. One self-published book somehow acquired PRH as an additional rights holder between the author filing a claim and the correction notices going out. One author found multiple titles claimed 100 percent by an individual they had never heard of who appears to be a scammer.

Sorting out a conflicting claim falls first to the claimants. The settlement provides a 30-day window for co-claimants to resolve differences between themselves. Until a difference is resolved, funds for that work are not released. Once the 30 days pass, the settlement administrator initiates contact to try to facilitate a mutually agreeable resolution. If that fails, the dispute goes to a court-appointed Special Master, whose decision is final.

What the Authors Guild recommends to authors comes down to paper. Submit a copy of the reversion letter or the contract to the portal; if no such document exists but the author is certain the rights reverted, submit that information anyway; and reach out to the publisher directly to advise that the rights are reverted and request that it amend the claim. The rule is already settled; the allocation turns on the evidence available to feed it, and that evidence is often a single letter from twenty years ago.

Milan set out what an author without that letter can offer instead. With the caveat that this is not legal advice, only what they would do themselves, the list runs to five: royalty statements that report on the books whose rights have not reverted but not on the ones that have, screenshots of the book's current publisher on Amazon, screenshots showing the book no longer appears on the publisher's current website, a notarized statement giving the date and the reason for the reversion, and an email to the publisher's royalties person asking them to confirm it. Milan adds that reversion letters are often sent to the agent, so the agent is the first place to ask. None of these five is a register. Each one reconstructs ownership out of circumstance.

Everything to this point is confirmed by reporting and primary sources. Seen from the data side, the shape of this is familiar. The records the industry did keep were built for selling and settling accounts, not for tracking who owns what: royalty statements, contract copies, a clause naming where the money should be sent. How agencies ended up in the rightsholder field says the same thing outright. A field recording where the money goes was held up against a question about who owns the rights.

Until now, who owned the rights to an out-of-print book was not a question with money attached to it. Nobody was paying for that answer, so nobody needed to maintain the record, and no system kept tracking reversions. Then the answer suddenly became worth $3,000, and 500,000 works needed it looked up at once.

Once the answer had a price, buyers for it turned up. On September 6, Strauss passed along an email authors had started to receive. A firm that buys up claims in bankruptcy and other distressed situations, describing itself as providing "immediate liquidity to creditors of distressed platforms," offered to purchase allocations in advance at up to $1,500 per book. Strauss wrote that such offers really aren't applicable to the Anthropic settlement, "where payout is probably only weeks or months away and the offer of 'up to' $1500 per book doesn't really improve on what many authors will collect anyway." The Authors Guild issued a separate warning about look-alike and misspelled domain names using the settlement as bait. When an amount is assigned before ownership is settled, corrections are not the only thing that comes through the gap. The last check on whether a claim is right turns out to be the eye of the claimant on the other side of it.

Music worked the same problem out in a register first. The Mechanical Licensing Collective is a nonprofit designated by the U.S. Copyright Office under the Music Modernization Act of 2018, and since January 2021 it has administered blanket mechanical licenses for streaming and download services, running a publicly accessible musical works database alongside a portal where creators and publishers submit and maintain their own works data. Book publishing has no equivalent register. An ISBN identifies an edition and says nothing about who holds the rights to that edition today.

None of this has any reason to stop at books. On September 5, The Seattle Times and Newsday sued OpenAI and Microsoft. Their complaint describes generative AI as "a snake eating its own tail" that could "destroy the very organizations" that produce the content it trains on, and argues that "AI products like ChatGPT and CoPilot are touted as producers of content, but in fact they are rapacious consumers, devouring human-authored content and delivering back to the world copies and derivative imitations of that same original content they consumed to achieve their commercial objectives." A Microsoft spokesperson told GeekWire that the company is "surprised by the lawsuit" but is "always happy to sit down and explore solutions to this type of dispute." News organizations have kept filing since The New York Times sued the same two companies in December 2023, and in July, Hachette, Cengage and Elsevier were among the publishers who brought a class action against Google.

If those suits end in settlements one day, the same question comes back around. Does this article belong to the publication or to the reporter, how do you count the version that ran in another outlet under a syndication deal, whose share is the freelancer's? Those answers have to be pulled out of documents, for works chopped much finer than a book. The Anthropic settlement is going through the preview version of that.

One question falls out of this for any organization that works with data. When the day comes that you have to pay for the data you use, can you pull out of your own records who to pay and how much? The answer needs three things written down: where the data came from, what terms it arrived under, and when those terms changed. In organizations that have handled training data provenance as a compliance matter rather than an accounting one, the third item is usually the empty one. Rights change, and the date of the change goes unrecorded.

Thank you for reading this far. The settlement terms and the author cases in this article come from TechCrunch's reporting, the categories of reports and the publishers' responses from Writer Beware, and the split rules and the dispute process from the Authors Guild guidance. If your team actually maintains the provenance of training data and the history of rights changes, we would be glad to hear what form you keep it in.

News Coverage

• 1.Ha, A. (2026). "Authors push back as publishers and agents seek share of Anthropic settlement." TechCrunch, Sept. 6, 2026.
• 2.Ha, A. (2026). "Seattle Times and Newsday are the latest publications to sue OpenAI and Microsoft." TechCrunch, Sept. 5, 2026.
• 3.Korosec, K. (2026). "Anthropic's landmark $1.5B copyright settlement is approved." TechCrunch, July 20, 2026.

Primary Sources

• 4.Strauss, V. (2026). "Anthropic Copyright Settlement: Publishers Are Making Incorrect Claims on Authors' Payouts." Writer Beware, Sept. 4, 2026.
• 5.Authors Guild. (2026). "Important Information Regarding Anthropic Copyright Settlement Claim Notices."
• 6.Milan, C. (2026). "On publishers not checking whether rights had reverted." Bluesky.
• 7.Milan, C. (2026). "Alternative evidence for proving reverted rights." Bluesky.
• 8.Milan, C. (2026). "On agents claiming a percentage of the settlement." Bluesky.
• 9.The Mechanical Licensing Collective. "About The MLC."

Industry Analysis

• 10.Business Model Analyst. "Anthropic Settlement: Why Proving Ownership, Not Pricing, Was the Hard Part."