Almost exactly a year ago, a group of authors filed suit against AI company Anthropic over its creation of an enormous library of digitized books to train its Claude LLMs. In addition to purchasing and scanning physical books (a la Google for its Google Books project), Anthropic also downloaded thousands of books that had been illegally uploaded to pirate sites, copying them multiple times for use in AI training.
Anthropic did not seek authors’ permission for any of this, arguing (as most AI companies do) that appropriating copyrighted material for AI training is fair use. The authors who brought the suit disagreed, arguing that Anthropic’s unpermissioned compiling and copying of their work constituted copyright infringment. (You can see the original complaint here.)
This past June, the judge in the case issued a mixed ruling, determining that while Anthropic’s digitzation of physical books, and its employment of its digitized library for AI training, both fit the four factors necessary to qualify as fair use, its downloading of pirated copies did not, and justified a trial on that issue. On July 17, 2025, the judge certified a class of LibGen & PiLiMi Pirated Books authors (LibGen is pirate site Library Genesis; PiLiMi is pirate site Pirate Library Mirror). The class consists of:
All beneficial or legal copyright owners of the exclusive right to reproduce copies of any book in the versions of LibGen or PiLiMi downloaded by Anthropic. “Book” refers to any work possessing an
ISBN or ASIN which was registered with the United States Copyright Office within five years of the work’s publication and which was registered with the United States Copyright Office before being
downloaded by Anthropic, or within three months of publication. Excluded are the directors, officers and employees of Anthropic, personnel of federal agencies, and district court personnel.
A more detailed overview of the ruling is here.
The judge has ordered the law firms representing the plaintiffs to provide names and street and email addresses of anyone who may be included in the class, so they can be contacted once the court approves an official class notice. One of the law firms representing authors in the suit has created a form authors can fill out to provide that information. Note that this does not add you to the class: it just makes it possible for you to be contacted when the official notice goes out.
The form, with an explanation, is here.
To find out if you potentially qualify for the class, you can search the LibGen database using this resource. I’m not aware of a similar resource for PiLiMi.
Remember, your work MUST have a US copyright registration per the quoted text above. This would seem to make it possible for non-US writers to join the class, as long as their works are registered in the US. Publishers are also eligible to join (relevant if your book is work-for-hire).
There’s currently a ton of litigation against AI companies for copyright infringement (along with a much smaller set of judicial decisions that are a mixed bag on the fair use issue). This is one of the cases that has gotten the farthest, and it is a big one–not just in terms of the stakes for authors and other creators, but in the sheer number of potential class members. It’s a potential extinction-level event for Anthropic, which could face billions of dollars in damages. Not surprisingly, it has appealed the class certification.
I’ll be keeping an eye on the case as it proceeds, and will update this post with developments as they happen.
UPDATE 8/26/25: Anthropic and book authors have reached a class-wide settlement. No details as of yet, but there will be a hearing on September 8 for preliminary approval of the settlement.
UPDATE 9/5/25: The details of the proposed class-wide settlement have been released. You can see the motion for preliminary approval of the settlement here. The details:
- Anthropic will pay a minimum of $1.5 billion into the settlement fund.
- Payment will be on a pro-rata basis, per work. With approximately 500,000 pirated titles (Anthropic downloaded many more pirated works than this, but only a subset of those books qualify–see the class definition above), that works out to an estimated $3,000 per work. The master list is still being finalized.
- Class members release their claims against Anthropic for its infringing actions re: the pirated books up to August 25, 2025. This release is qualified in three respects: it extends only to past claims and does not bar claims for new infringement after that date; it pertains only to works on the master list (i.e., if you’ve written five books and three are on the list, you release your claim for those three only); and there’s no release of any claims stemming from infringing outputs from Anthropic’s AI products.
- Anthropic will destroy the two pirated databases (LibGen and PiLiMi) after all litigation has closed.
I tend to be a pessimist, and was expecting the payout to be under $100. $3,000, obviously, is a lot more than that. It’s qualified, though, by the fact that lawyers’ fees (25%, according to the motion), expenses, and $50,000 awards for the class representatives (the authors whose names are on the suit) will be deducted from the fund before payouts are calculated. Additionally, publishers are eligible to be class members, so if your book is currently in print under an exclusive contract, the payout may be shared with your publisher.
The Authors Guild has a detailed explainer on the settlement.
There’s now an official settlement website where you can provide your contact information to the class counsel (this does not add you to the class; it just makes it possible for you to be contacted with notices about the case). The court must allow time for objections and comments, so don’t expect things to resolve quickly.
The settlement is being hailed as a massive victory for authors (apparently it’s the biggest copyright infringement recovery in history) and as a turning point for AI companies. However, as a friend pointed out to me, the settlement isn’t really about AI training at all. Rather, it’s about book piracy. Anthropic is being punished only for the method by which it obtained some of the books it used for AI training. The training itself was deemed by the judge to be fair use.
Also…$1.5 billion is a huge amount of money. But consider that Anthropic just raised $13 billion in its latest fundraising round, and the company itself is valued at $183 billion.
It’s early days in the AI training copyright wars. According to this AI copyright lawsuit tracker, 48 cases have been filed as of mid-August, and there are certainly more to come. Stay tuned.
UPDATE 9/8/25: The judge in the case just rejected the settlement, with some pretty salty language.
If AI training itself is fair use, but the method of acquisition is the problem, does that mean future lawsuits will focus less on training and more on sourcing?
Sourcing is important in this particular case, but it won’t necessarily be in others. This is also just one ruling in just one case–there are something like 48 others ongoing at present, and there will be more. So the fair use queston is very, very far from being resolved (though I do fear that it’ll ultimately be resolved in AI companies’ favor). In any case, the main concern for creators is copyright: the fact that AI training on works without permission–regardeless of how they are sourced–is copyright infringement. That’s the basis on which most cases have been filed, and I think it will continue to be so. Sourcing really is a side issue, IMO.
I got my novel copywritten for protection, because I sent a hardback copy to a SCAMMER Publisher/Marketer and they can use my idea to make money. I feel better knowing that I am safe and if they use my idea, it will be easier to sue them. Believe me, when I say I will SUE! This is my novel and not theirs.
Unfortunately ideas aren’t protected by copyright–only the expression of those ideas (the actual written words). When you say you got your novel copyrwritten, did you register your copyright with the US Copyright Office or the intellectual property office of another country that has an official registration process? (By law, you own copyright from the moment you write down the words.)
The US copyright stipulation seems to be another loophole designed to eliminate many authors from the case. There is no requirement in the UK – as soon as you create, the work is yours. Four of my books are on the database, I have ISBN numbers for all, I sell books in the US and there is a copyright notice on all of my books – and yet, according to the rules, nope, not entitled to anything because didn’t register copyright in the US. An objection should be made to eleminate this US copyright rule.
All of my books are on the pirate site. I’ve read and seen (in videos) that only books registered for copyright through Aug. 10th, 2022 are included in the settlement. Do you know if this date is accurate?
I don’t know the exact date, but they must have been registered before Anthropic downloaded them. So that might be the download date, but I don’t know for sure.
Thank you for your service, Victoria!
Hi Victoria, thank you for posting the details on this. My works qualify for the class, and I want us to refuse this offer and fight for more. Are we going to get a chance to ask the firm to refuse this offer and take it to court?
As I understand it, there will be an opportunity to offer comment. Details on that, as well as on how to file a claim, will be provided via notices from the settlement administrator, so it’s important to fill out the form on the settlement website to provide your contact information.
Also as I understand it, finding your works in the LibGen database doesn’t necessarily mean they’re part of the class. Plaintiffs are working on the final list, and it won’t be complete until sometime in October. Authors will need to check it to find out if their books are included.
Thank you so much for the info and links; I had a bunch of stuff listed on LibGen. For those looking to see if anthologized works are listed: You can search by editor name. I had more than a few stories & novellas listed. I would hope that some type of pro-rata payment might be forthcoming. I als recommend searching LibGen directly–but use a VPN.
According to Torrent Freak, a piracy blog, it appears the team behind Pirate Library Mirror moved the work over to another website called Anna’s Archive. The website specially states they are a book and paper scraper for LibGen as well as other sites, claiming they are open source. I searched for my book and found it on their site with download statistics. While my book didn’t initially come up in the Atlantic link, my name did, leading me to investigate. I hope this info helps other authors in finding out if their book was pirated: https://annas-archive.org/
What is your opinion about this: Why hasn’t Writer Beware written about the SFWA scam?https://www.quora.com/Why-hasnt-Writer-Beware-written-about-the-SFWA-scam
I had to roll my eyes at that quora post, which is an obvious hit piece on me and Writer Beware, written by someone with a grudge (a scammer, maybe?). At any rate, I couldn’t have improved upon this detailed post at File 770, which examines these year-old events in a more balanced fashion.
I’ve found both my novels and one non-fiction book.
is there a method for searching the database for inclusion as an author in an anthology or should the search be for the title of the book?
The Atlantic Search function only looks for names, not titles, but it does include some anthologies under the editor and contributor names.
I found two editions of one of my textbooks, and 40 issues of The Grantville Gazette
Can anyone explain the legal reason why a book has to be registered with the Copyright Office? It’s my understanding that all works are automatically copyright by the author, and that putting a copyright notice on the work is more of a formality so that a thief can’t just say, “Well, I didn’t know!” Registering with the copyright office isn’t supposed to actually change the strength of ownership, only make it even more clear. So why did the judge put copyright registration as one of the requirements to be a member of the class for the class-action lawsuit?
https://copyrightalliance.org/faqs/why-register-copyright/
In the USA (and only the USA; other countries don’t have this requirement), copyright registration is a pre-requisite for taking legal action to defend your copyright. That’s why you or your publisher have to have registered your copyright in order to qualify for the class.
It has 22 of my 29 books and all are officially copyrighted works. So this company already has them — May these AI companies pay whatever price they must for their pirating of books — as you and I would pay had we done this. They should get no special treatment.
I have 169 hits on that, but I don’t have my books registered for copyright. Oh goody.
You can register at any time during the life of copyright (your lifetime plus 70 years).But for registration to count as prima facie evidence of copyright ownership (important if you ever go to court, because it saves you having to prove that you own the copyright), you must register within five years of publication. If you register after that date, it’s up to the court to decide if registration proves ownership or not.
The AI companies must pay for usage just as a patron pays to read a book or copy write material. Even libraries buy their books. There is a cost to write material and to publish that material and distribute the material. If the AI companies don’t want to pay, they should write their own material and use that material for training computer chips, software and storage databases.
One of my books is in the LibGen database. It’s not fiction, is it likely to have been used to train AI?
Yes–Anthropic used both fiction and nonfiction books.
I don’t suppose there is any method to get your books removed from the pirate databases, huh? I was surprised to find ALL of my novels in the pirate site’s library. (Grrrrr)
Some pirate sites include intellectual property or DMCA pages, but these are mostly just window dressing; there’s nothing to compel them to actually take down a book if you contact them, and they are likely just to ignore your request, or fake-remove the book so that you (or rather your browser) can’t see it but to others it remains perfectly visible. Really, group actions that close the site down or hit it with heavy penalties are the only reliable way to remove content.
If you’re trad-pubbed, you might consider contacting your publisher to let them know about it. And it never hurts to send a notice, or fill out a form–just in case the site pays attention. Don’t hold your breath, though. In these situations, individual authors options are limited to non-existent.