The $3,000 Check Has Strings: Authors and Publishers Fight Over Anthropic's $1.5 Billion Book Settlement

A year after Anthropic agreed to pay $1.5 billion to settle the biggest copyright case in American history, the money is finally moving — and the fight has moved with it. This week, the settlement administrator began sending claim notices to every author and publisher who filed in Bartz v. Anthropic, listing each claimed title, who else claimed it, and — crucially — what percentage of the award the other parties say they are owed. Within hours, authors’ forums, the Authors Guild, and the anti-scam watchdog Writer Beware were fielding a wave of complaints: publishers and literary agencies are claiming slices of the payout that many writers say they never agreed to.

The New York Times, which reported the standoff on September 5 under the headline “Authors Wrangle With Publishers Over $1.5 Billion Anthropic A.I. Settlement,” captured the mood in the books world: relief that the checks are coming, dread that someone else may be holding the scissors.

A very brief recap of how we got here

Bartz v. Anthropic was filed in August 2024 by novelists Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson, alleging that Anthropic pirated their books — along with roughly seven million others — from shadow-library datasets LibGen and PiLiMi to train its Claude models. In June 2025, Judge William Alsup of the Northern District of California delivered a split landmark ruling: training on legally acquired books is fair use, but downloading pirated copies is not. Facing a December 2025 piracy trial, Anthropic settled for $1.5 billion — approximately $3,000 per eligible title across some 500,000 books that survived deduplication and eligibility screening (the works had to carry an ISBN or ASIN and have timely U.S. copyright registration). Final approval came in the summer of 2026, claims closed on March 30, and Anthropic has so far deposited $1.05 billion into escrow — an account that has already earned nearly $25 million in interest, with another $450 million-plus still to come.

What the notices revealed

Per a status report by class counsel, first payments should land between November 1 and November 15, 2026 — but only for “works where all rightsholders are in agreement and where all necessary payment information is provided.” The first payment works out to approximately $2,203.56 per title in gross proceeds, after deduction of fees, service awards, and expenses.

That per-title pot is where the trouble starts. The settlement’s structure is simple on paper:

• Sole rightsholders — self-published authors, or writers whose rights have reverted — are entitled to 100% of the roughly $3,000 award for each title.
• In-print trade and university-press books carry a 50/50 default split between author side and publisher side, reflecting standard contract norms. Co-authors and co-publishers divide their share.
• Educational and textbook titles have no default; claimants make good-faith representations from their contracts.
• If claimants can’t agree, the split goes to a court-appointed special master to adjudicate.

The default mechanism was designed to make the process fast: check a box, split evenly, no contract review needed. But the notices exposed how often reality deviates from the default — and how aggressively some intermediaries moved to claim more.

“Five authors, six agencies, 15 to 25 percent”

Writer Beware’s Victoria Strauss, whose September 4 post catalyzed much of the discussion, reported hearing within a single morning from five authors whose six different literary agencies had made claims for between 15% and 25% of the allocation on books they agented. There is no basis in the settlement for an agency cut of that size as a default; commission rates on subsidiary or incidental income vary by agency agreement, and authors who never negotiated “legal settlement proceeds” as a commissionable income stream are now discovering their agency expects a slice anyway.

The publisher-side behavior has drawn sharper fire. Multiple authors with reverted rights — books whose contracts terminated, returning all rights to the author — found their former publishers nonetheless claiming 50% or in some cases reportedly up to 100% of the award on titles they no longer control. The Authors Guild’s updated guidance notes that rights-reversion cases are a known source of erroneous publisher claims, and Penguin Random House’s own settlement FAQ stresses that class members must each file their own claims and that publishers cannot file on authors’ behalf. Small-press authors have reported disputes over split amounts across dozens of titles at a time.

What authors should actually do

The practical advice converges on a few points. First, read the notice carefully: it lists every co-claimant and their claimed percentage for each title — that document is the only authoritative map of who wants what. Second, check your contract before conceding anything: if your publishing agreement terminated or your rights reverted, you are the sole rightsholder and should contest the publisher’s claim. Third, don’t panic-opt-out of a negotiation: claimants who reject the default are asked to state their good-faith position, and unresolved disputes escalate to the special master — slower, but far better than silently absorbing an incorrect split. The Authors Guild is assisting members through its legal request form, and the settlement site’s FAQ walks through the disagreement process.

Timing matters. Only agreed splits get paid in the November 1–15 window; everything contested rolls into later disbursements. There is a real tension between disputing a greedy claim and seeing money this year.

Why this matters beyond the check

The Bartz settlement was supposed to be the clean ending to the first chapter of the AI-copyright wars: a lab caught pirating pays record damages, destroys its LibGen/PiLiMi corpus, and the industry moves on. Instead, the endgame has revealed that the hardest part of compensating creative work at scale isn’t liability — it’s allocation. Five hundred thousand titles, millions of counterparties, decades of heterogeneous contracts, agencies and publishers and co-authors all holding partially overlapping claims: the settlement has become a stress test of whether the publishing industry even knows who owns what.

That lesson is now being watched closely by every rights holder negotiating with AI companies. OpenAI, Microsoft, Meta, and others face a rising tide of copyright suits — including a new one from the Seattle Times and Newsday just this month seeking destruction of training datasets. If those cases settle too, the machinery built for Bartz — the works lists, the default splits, the special masters — is the template. The authors currently fighting their own publishers over 15% of $3,000 are, in effect, debugging that machinery in public.

The escrow account, meanwhile, keeps earning interest. Someone will get paid. The question this week made vivid is exactly who — and how much of the largest copyright recovery in U.S. history survives the trip through the contracts that the AI industry, ironically, pirited in the first place.