Okay, I have to tell you something about this Anthropic settlement that I think most people are going to miss, and it matters more than the $1.5 billion headline.
If you have been following the news at all, you know what happened. Anthropic, the company that makes Claude, got caught downloading a lot of books from a pirate library called LibGen and using them to train their AI. They did not pay. They did not license. They just took. Then they got sued by a class of authors, and in September they agreed to pay $1.5 billion to settle. The biggest copyright settlement in U.S. history.
For most people that is where the story ends. Good for the authors, right? AI company finally pays for what it took.
Here is the part I have to tell you about. The $1.5 billion gets split among the books on what they are calling the Works List, and that list has 482,460 entries. If your book is on the list, you get about $3,000. Sounds reasonable. Except the Works List was built in a way that systematically undercounts indie authors who used what is called group copyright registration. Which, if you do not know, was the Copyright Office's own recommended way for prolific writers to register their books for almost twenty years.
I am one of them. So is something like 600 to 5,000 other indie authors. We are about to find out, when our checks arrive, that our registration choice, the one the Copyright Office encouraged, is the reason we are getting one share when our peers who registered each book separately are getting forty.
Let me back up and show you how this happened, because honestly it is the kind of thing that has been making me crazy for eight months.
What Is Group Registration, Plain English
Before March 15, 2019, if you were the kind of author who finished a book every couple of months, and romance is a fast genre so I was one of them, you could register a whole batch of books with the Copyright Office in a single filing. They called it an unpublished collection. It was cheaper, it was faster, and it was the path the Copyright Office literally walked you through on their website. Trust me, nobody was doing this to game the system. We were doing it because that was the form.
I have a registration that covers more than forty of my novels. Each one has its own ISBN. Each one has its own title. Each one was published as its own book, sold as its own book, read as its own book. But on the Works List in this settlement, all forty-plus of them show up as one entry. One $3,000 share.
So an author who registered each of her forty books one at a time gets $120,000. I get $3,000. Same books, same theft, same scope of pirating by Anthropic. Different share by a factor of forty.
The Math on How There Are 5,000 Authors Like Me
Honestly, this is the part I want to walk you through carefully because the number sounds enormous until you do it step by step.
There are 482,460 entries on the Works List. The Copyright Office itself, in writing, in a 2019 federal register notice, said that pre-2019 group registrations could cover dozens, hundreds, or even thousands of works per filing. So a chunk of those 482,460 entries are not one book each. They are registrations covering many books.
Nobody outside the Settlement Administrator knows the exact share. But you can build the estimate from public data. Indie authors make up roughly a third of U.S. ebook sales by volume. The genres that used group registration most heavily are romance, mystery, sci-fi, fantasy. Basically all the indie-dominant categories. Authors in those genres typically published anywhere from 15 to 40 books in the years group registration was available.
If even 5 percent of the 482,460 Works List entries trace to a pre-2019 group registration, that is about 24,000 books being counted as one share each instead of one share per book. At 15 percent, which is plausible given how heavily indie authors leaned on this option, it is 72,000 books.
Divide that by 15 to 40 books per registration and you get how many authors are affected. Low end: 24,000 divided by 40 is about 600 authors. High end: 72,000 divided by 15 is about 4,800 authors. Round up, call it 5,000.
So somewhere between 600 and 5,000 indie authors are about to find out that the registration choice the government encouraged is going to cost them most of their recovery.
Here Is the Part That Has Me Frustrated
This was solvable. I have to tell you, I keep coming back to this part, because it is so technically easy.
A group copyright registration is a PDF. The Copyright Office issues it electronically. It lists every single title covered, in plain text, on the certificate. Modern PDFs can be read by a computer in about three lines of code. The Settlement Administrator in this case already runs sophisticated matching across ISBN databases, Amazon records, Copyright Office records, and the LibGen book index Anthropic was forced to produce in discovery. Adding a step that says, if a registration is a group registration, pull the title list off the certificate and match each title against the LibGen index, is in engineering terms a Tuesday afternoon.
I sent the lawyers my registration certificate in November 2025. They had it. They had access to LibGen. They had the technical capacity. They had eight months between when I raised it and when the final approval hearing happened. And honestly, they chose not to do the work.
I am not being mean when I say that. I am saying it because it is what happened. When you have the data, the tools, the time, and the technical capacity to fix a known problem and you decide not to, that is a choice. It deserves to be named as one.
There is a rule for this too. It is called Rule 23(g)(4), and it is part of the federal rules that govern every class action lawsuit in federal court. The rule says class action lawyers have to fairly and adequately represent the interests of the class. All of the class. Choosing not to fix a methodology that excludes a subgroup, after the subgroup has raised it for eight months, is exactly the kind of thing that rule was written to address.
The Lawyers' Cut and Why the Dilution Worry Does Not Hold Up
Okay so this is where the math gets a little hot, but stay with me because it is honestly the part nobody is talking about.
If you read the news on this settlement, you might worry: well, if the court counted my forty books separately, every other author on the list gets a smaller share. The fund is fixed. There is no extra money. Right?
Wrong. The fund is fixed only after the lawyers take their cut. And the lawyers are asking for a cut that the court itself, in the proposed order, has called on the high side.
Class Counsel, the firms representing the class, are asking for $187.5 million in fees. That is 12.5 percent of $1.5 billion. They originally asked for 20 percent, which would have been $300 million, and the court pushed back. They came down to 12.5.
But here is the thing. The way 9th Circuit courts evaluate whether attorneys' fees are reasonable in class actions is by comparing the requested fee to the lawyers' actual hours. Their hours times their rates is called the lodestar. The ratio between what they are asking for and their lodestar is called the multiplier. The 9th Circuit treats multipliers between one and four as presumptively reasonable. Four to eight is exceptional. Above eight is extraordinary.
Class Counsel's lodestar in this case, including actual hours plus projected work through 2027, is about $27 million. Their $187.5 million fee is a 6.92 multiplier. That is the high end of exceptional, and the court has noticed.
The judge announced from the bench at the May 14 hearing that she will conduct her own independent lodestar review rather than taking the lawyers at their word. The proposed order holds back 10 percent of the fee, $18.75 million, conditional on the lawyers' actual time matching their projection. She has been questioning their money for months.
Here is what that means for the dilution worry. If the multiplier came down from 6.92 to 4, still inside exceptional, the lawyers' fee would drop from $187.5 million to about $108 million. That is $80 million back in the fund. At $3,109 per work, that is enough to pay full per-work shares for more than 25,000 additional book-units without diluting anyone.
If the multiplier came down to a clearly reasonable 3, the recovered amount approaches $107 million. That is enough to fully count every group-registration author at the conservative subgroup estimate, no dilution to anyone.
Here is who that money would actually pay. Every one of those 600 to 5,000 indie authors met every requirement Class Counsel themselves wrote into the settlement. Properly registered with the Copyright Office. Registered in the timing window. Registered before Anthropic downloaded the book. ISBN-bearing. On LibGen. They check every single box the lawyers themselves drafted. They are not asking to be added to the class. They are already in the class. They are being undercounted because the lawyers picked a methodology that excluded them and then declined to engage when the exclusion was identified.
The dilution worry is about how much the lawyers want, not about whether the affected authors deserve to be counted right.
What I Actually Think Will Happen
Here is the realistic prediction, because I am not going to pretend otherwise.
Most legal scholars who watch class action settlements will tell you, honestly, that judges in mega-settlements tend to sign the deal the lawyers brought them. There are practical reasons. The judge has fifty other cases on her docket. The settlement has years of work in it. The claims rate is over 92 percent. That is overwhelming class approval. The path of least resistance is to sign. That is what is most likely to happen here too.
But.
Copyright is the default rule. The Constitution gives Congress the power to protect authors. The Copyright Act says infringement is unlawful. Fair use is a narrow exception, and in this case the court already ruled that downloading the LibGen and PiLiMi datasets was not fair use. The default, that authors get protected, applies.
There are also actual rules about how a class action settlement has to be evaluated before a judge can approve it. Federal Rule of Civil Procedure 23(e)(2) lists four of them. Two of those are directly on point for what I am describing.
The first is Rule 23(e)(2)(A). It says the class representatives and class counsel have to have adequately represented the class. Adequate means actually engaging with the issues that affect different parts of the class. Not just the parts that fit the chosen methodology.
The second is Rule 23(e)(2)(D). It says the proposed settlement has to treat class members equitably relative to each other. That rule was added to the federal rules in 2018 specifically to make courts pay attention to settlements that treat similarly situated class members very differently. Two indie authors who were both pirated the same way, who both meet every registration criterion the settlement requires, should not receive shares that differ by a factor of forty because of how their copyrights were filed administratively.
And the 9th Circuit, the federal appeals court that covers California where this case is being heard, has a case called Briseño v. Henderson. In Briseño the court said that when an objector raises a substantive issue, the trial court is not allowed to just say overruled and write boilerplate. The judge has to actually engage the argument. Not approve the settlement and pretend the objection was not raised.
These are not exotic rules. This is law school 101. I am not asking for something new. I am asking the court to apply rules that already exist.
When a settlement administrator builds a Works List that mechanically excludes properly registered, properly timed, ISBN-bearing, individually published books because they happened to be registered together, the settlement has departed from the default. It has also departed from Rule 23(e)(2)(D). The default, and the rule, would have been to count each book.
And I am not the only objector saying this. Of the 53 objections filed in this case, a substantial fraction make some version of the same complaint. The methodology systematically cut out independent authors and independently published books that meet every substantive requirement the lawyers themselves wrote into the class definition. Here are the examples.
Werner objected because her publisher failed to register her books, even though she meets every other criterion of the class.
Chase objected because the pseudonym matching was inadequate. Authors who wrote under pen names but whose real names were on file with the Copyright Office got cut by the matching algorithm.
Smith objected because two of her qualifying books are missing from the Works List. She has the registrations. The books were on LibGen. They were left off.
The Italian publishers, eight separate publishing houses including Newton Compton Editori, Tea Tascabili, Casa Editrice Nord, Bollati Boringhieri, Adriano Salani Editore, Longanesi, Antonio Vallardi, and Garzanti, all objected on the same ground. Their qualifying works were excluded because the lawyers' methodology defaulted to U.S.-only matching.
Watkins Media Limited and Collective Ink Limited, two UK publishers, objected on the same systematic-exclusion ground.
Burton and DesJardins, co-authors, objected because the author-publisher split methodology was built in a way that disadvantages independent and self-published authors.
Hampton and Chakanga objected because the notice itself, and the procedures for objecting, were not adequately explained to class members who were not lawyers.
And I objected because my pre-2019 group registration covering more than forty individually published books shows up as one share instead of forty.
Different facts. Same defect. Across more than a dozen separate objections, by class members in different countries, in different genres, with different stories, the same structural pattern emerges. The lawyers built a settlement that systematically cut out independent authors and independently published books that fit every substantive criterion of the class. They cut them out at the administrative layer. Then, when the affected class members raised it, the lawyers declined to engage.
Some of those objectors are wrong on the law. Copyright registration genuinely is required. Unregistered books are not class works. But even setting those aside, the pattern is the same. People who meet the substantive class definition, who check every box the lawyers themselves drafted, got cut by the methodology the lawyers chose, and then were told the methodology was final.
That pattern is what the court has been looking at while reviewing the proposed order. Even if the judge signs the deal, and the realistic prediction is she will, with or without modifications, the record needs to reflect that the pattern existed and was named in writing. The next AI copyright settlement is being drafted somewhere right now. The lawyers will read this one. The objectors in the next case will cite the objectors in this one.
The argument deserves to be on the record even when the argument is probably going to lose. Copyright is the default. The default deserves to be argued for.
If You Used Group Registration Too
Let me be honest about where this goes from here.
Start with your registration certificates. If one of them lists multiple titles, that is a group registration. If those titles show up on the Bartz Works List as a single entry, you are in the subgroup I have been describing.
A small bit of background that matters. I went to law school years ago. I had a civil procedure professor who drilled Federal Rule 23, the rule that governs every class action, all year. He thought I was not paying attention. To be fair, I once missed a class because I had a day off and I flew to Ireland for it. He really tried to throw the book at me for that one. Trust me, the rule stuck.
So when the lawyers in this case told me my request for proper counting of my forty-plus books was a windfall, that got my Irish up. Because it is not a windfall. It is the rule. Rule 23(e)(2)(D) was added in 2018 specifically to stop courts from approving settlements that treat similarly situated class members very differently. Calling that windfall is not how the rule works.
Here is what is realistic.
I cannot represent anyone but myself in this case. I am pro se. I am speaking, not representing. But I am speaking because this truly does affect others, and the others have to know what to do regardless of how my objection lands.
Path one. If the judge rules for the exception I asked for, then everyone in the group-registration subgroup gets to file a claim dispute through the court-appointed Special Master. The mechanism is in the Plan of Allocation, section 3(d). The documentation is your group registration certificate. The argument is the class definition itself, which is defined by books with ISBNs, not by registration numbers.
Path two. If the judge does not rule for the exception, I am preparing for an appeal. The record is preserved across my filings. The rule is the rule, and the 9th Circuit can apply it.
Path three, and this one is independent of my objection. There is a clear case for any author whose qualifying book is not on the Works List at all to file a claim for that unpaid work. The settlement does not release anything that is not on the Works List. If you have a registered book with an ISBN that was downloaded from LibGen or PiLiMi and it is not on the Works List, you retain every right to bring your own claim against Anthropic for that book. The settlement does not touch those books. That right is independent of whatever happens with my objection.
To be ready for any of these paths, here is what to gather now. Your group registration certificate, the PDF from the U.S. Copyright Office that lists every title covered. The full ISBN list for every book on that certificate. Documentation that each of those books was on LibGen or PiLiMi, which you can verify through The Atlantic's searchable LibGen database. If you have books that were not on the Works List at all, gather their registrations and ISBNs separately, because those are path three.
You are not alone. The Copyright Office documented this subgroup in writing in 2019. The Authors Alliance has been writing about it. The academic literature on AI copyright has flagged it. We are a recognized subgroup with a documented record. The question is whether one federal judge in San Francisco is willing to apply rules that already exist to a problem the federal regulator already named. The answer to that question is what we are waiting for.
The Last Thing
If you are a reader of mine: thank you for being here. The books that taught Claude how to write love stories are the same books you have been reading. They came from work I did at a real desk in a real chair, registered with the Copyright Office, sold to people who wanted them. That work is mine. It is still mine. We are arguing about how it should be counted in this settlement. That is one piece of a bigger fight about whether AI companies have to actually license what they use.
Both fights matter. I will write again when the judge rules.
Victoria Pinder
