⚖️ Bartz v. Anthropic: what authors need to know about payments in 2026

In summer 2026, a dispute that had kept the book market on edge for three years reached its final point: a San Francisco court confirmed the largest payout in US history in a copyright case. For writers, this is a rare instance where an abstract lawsuit about training language models turns into a concrete transfer to your account, and also a reason to take a fresh look at publisher contracts and external book promotion channels, including a platform for guest posts. In this piece, we will break down what the court decided, who gets what and how much, why the amount you actually receive differs from the widely circulated "three thousand dollars per book," and what to do if you were not included in the distribution.

What an author should do after final approval

💡 Quick overview:

• Step 1: Make sure your titles are on the court-approved list of works and that your claim was filed before the deadline.
• Step 2: Agree with co-authors and the publisher on the share assigned to each party under the current contract.
• Step 3: Check your payment and tax details with the claims administrator: outdated information delays transfers more than any legal dispute.
• Step 4: Keep your claim confirmation and correspondence with the publisher; they will help if there is a disagreement about splitting the amount.
• Step 5: Monitor the appeals track: as long as objections remain open, distribution of the fund will not start.
• Step 6: Decide what to do with your rights going forward, from the terms of new contracts to participating in separate proceedings.

The sequence sounds dull, but this is exactly where money gets lost. The claims administrator works with hundreds of thousands of entries and does not track down rights holders manually: if your details are outdated or the share is not agreed, the payment goes into a deferred queue and sits there for months. Below is a short video breakdown of the ruling from an international news agency, followed by details on each point.

From pirate archives to a record fund

The timeline matters more than the emotions here. The lawsuit was filed by three authors of nonfiction and genre fiction, and the claim concerned not the fact of training the model itself, but the way the texts were obtained. In June 2025, District Judge William Alsup issued a split ruling: he found that training on legally purchased books was fair use, while storing copies downloaded from pirate sources fell outside that protection, as the Authors Guild covered in detail in its analysis of the interim ruling.

That summer, the court certified the class action, but narrowly: only for the pirate copies episode and only for books with timely registration at the copyright registration office and an assigned publisher number. The disputed files came from two major shadow book archives, Library Genesis and Pirate Library Mirror, from which, according to the case record, roughly seven million copies were downloaded; after removing duplicates and ineligible titles, about 482,000 works ended up in the class, as noted in the legal analysis of the settlement structure.

The court granted preliminary approval of the settlement on September 25, 2025, and Judge Araceli Martinez delivered the final word: on July 20, 2026, she approved the agreement definitively and issued the final judgment in the case, as reported by the Authors Guild in its news item on the final approval. The fund size, $1.5 billion plus interest, made this settlement the largest in the history of American copyright law, as also noted in a report by the legal publication JURIST.

How much the rights holder will actually receive

The figure of about three thousand dollars per work that spread across news feeds is not the author's fee, but the amount for one edition as a whole, as noted in the overview of the fund distribution terms. It is divided among everyone who holds rights to a particular book. Then the share scale applies: by default, trade and university contracts assume an equal split between the author and the publisher, co-authors divide the author's portion among themselves, and multiple publishers with rights to different formats divide the publisher's portion.

The default option was chosen for speed: if both parties agree to it, no contract needs to be presented and the claim follows a simplified route. If the author or publisher rejects the default, each party states its own good-faith estimate of the share, and unresolved disagreements go to a court-appointed arbitrator. The practical takeaway is simple: arguing about percentages is cheaper before filing a claim than after it.

Another factor that pushes the final payout upward is the share of unclaimed items. By April 16, 2026, rights holders had claimed 440,490 works out of 482,460 on the list, or 91.3 percent, and the court called this activity abnormally high for class actions, as noted in the summary of submitted claims. The smaller the undistributed remainder, the closer the actual amount comes to the original estimate.

Where part of the fund goes: fees, expenses, and the payment schedule

The fund is structured without clawback: everything not spent on administration and lawyers stays with rights holders rather than returning to the defendant. The main expense item turned out to be noticeably more modest than expected. Class counsel initially requested about 300 million dollars, then lowered the request to 187.5 million, and the court awarded 101,561,111 dollars, roughly seven percent of the fund, applying a 3.75 multiplier to the lawyers' base rate of about 27 million instead of the requested one, and these calculations are provided in the detailed breakdown of the settlement terms.

The money does not arrive in a single transfer. The first tranche of 300 million dollars was paid by October 2, 2025, and the remaining payments are scheduled through September 25, 2027, a timeline also reflected in the chronology of settlement contributions. This explains the cautious wording about timing: distribution begins after appellate options are exhausted, not on the day the decision is signed.

The per-work rate itself came out roughly four times higher than the minimum statutory compensation for willful infringement, an unusually generous result for a class action, as emphasized by a technology publication in its approval review. For comparison: in most consumer class actions, participants receive token amounts, while here the money is comparable to an advance for a small book.

Why lawyers don't call this a win for authors

The money has been received, but there is no precedent. The settlement closes a specific episode involving pirated copies and does not create a binding rule for other disputes, while the interim finding on fair use in training on legally purchased books has not gone away. This duality is exactly what the public broadcaster's analysis of the limits of success emphasizes: authors won compensation for the method of obtaining the texts, not recognition that training a model on a book requires a separate license.

The practical takeaway for the market is unpleasant: by purchasing the same corpus legally, a developer follows the same logic with no payments at all. That is why parallel cases continue on their own track. The lawsuit against Meta over training its language model remains alive with respect to file sharing via torrent downloads, and the overall litigation landscape is laid out in detail in an industry review of case law.

Those who deliberately opted out of the class are a separate story. A group of authors, including a two-time Pulitzer Prize winner, declined a share of the fund and filed individual lawsuits against several developers at once, seeking $150,000 per work from each defendant under the statutory damages cap, as reported by a book industry trade publication. The strategy is risky: an individual case drags on for years and requires money for lawyers, but the ceiling on claims is incomparable to a share in a common fund.

A real-world example: three books, three different outcomes

Consider a composite situation that recurs in author correspondence dozens of times. A writer has three published books. The first came out at a major publisher under a standard contract, and the rights are active: the payment for it is split evenly, with half going to the publisher. The second book is old, the contract has expired, and the rights have reverted to the author: here the author files the claim alone and keeps the full amount. The third was co-authored and published by a university press as a textbook: the default scale does not apply to such editions, so the shares will have to be justified by the contract language, and if there is a disagreement, the dispute will go to an arbitrator.

The difference in outcome arises not from the quality of the books, but from the state of the rights at the time of filing. That is the main practical lesson: a registry of your own contracts, termination dates, and format licenses should be maintained continuously, not reconstructed a week before the deadline. The same registry will be useful in the next class actions, and judging by the pace of litigation, they will certainly come.

What to do if you did not file a claim

A missed deadline does not always mean lost money. Administrators of large funds usually run additional rounds for unclaimed items, and if a significant balance remains, the court may authorize redistribution among already confirmed participants. Check the status of a specific work on the official administration website, not through retellings on social media: fake forms that collect personal data under the guise of a claim regularly surface there.

If a book did not make the list at all, for example because rights were not registered in time, participation in this fund is impossible. A sensible response for the future: register rights to new editions without delay, include a separate clause in contracts covering the use of texts for machine learning, and preserve evidence of authorship. A general reference on the class composition and payment procedure is kept current in material from a professional authors' association, and it is worth checking whenever the case status changes.

⁉️🤔 Frequently asked questions

When will authors receive the money?

No exact date has been announced. Distribution starts after appeal deadlines expire and filed objections are resolved, so the realistic target mentioned by class representatives is late autumn as the earliest possibility. The administrator will notify claimants by email at the address provided in the form; separately requesting status is usually unnecessary.

Is it true that each author will receive three thousand dollars?

No, that is the amount per work before expenses are deducted, and it is divided among all rights holders of the book. An author with an active publishing contract receives half, and co-authors split the author's portion among themselves. Only someone who holds the rights solely, for example in self-publishing or after rights reversion, receives the full amount.

Did the court rule that training neural networks on books is illegal?

No. The interim ruling expressly called training on legally acquired copies fair use. The claims and the money relate exclusively to downloading and storing pirated copies, so the settlement does not prohibit developers from training models on purchased books and creates no such prohibition for the future.

What happens to unclaimed fund shares?

The fund does not revert to the defendant under any scenario. The parties anticipate redistributing the remainder among confirmed participants, and if it proves too small for an economically sensible distribution, it will be directed, with court approval, to charitable purposes related to supporting writers.

Is it worth opting out of the class for an individual lawsuit?

Opting out of this case is no longer possible; the deadlines are closed. Those who opted out earlier face an independent proceeding with legal fees and the risk of losing on the fair use question. For most authors, a guaranteed share in the fund proved more rational than a multi-year dispute with an unpredictable outcome.

How will this settlement affect new publishing contracts?

It creates no direct obligation, but it sets a market benchmark for the value of text used to train models. Lawyers advise putting machine learning licensing in a separate contract clause rather than leaving it inside a general electronic rights provision; otherwise the interpretation will fall to the publisher.

Takeaways: what to carry away from this story

The record fund has been confirmed by the court, the share of claimed works turned out to be unusually high, and the rules for dividing the amounts are known in advance. At the same time, the legal framework remains unchanged: the payment is for the method of obtaining the texts, not for the fact of training itself. That means the next wave of disputes will be built around licenses and contract language, not around loud accusations of piracy.

What to do next if you write and publish: check the status of your claims and payment details, put your rights registry in order across all editions, add a separate clause on the use of text for model training to your new contract template, and do not ignore emails from administrators of such funds. If this material was useful, take a look at our other explainers on copyright and digital rights: the topic is changing fast, and the next court ruling will once again redraw the balance of power.