Did you know your book could already be part of an AI training dataset?
Maybe it was downloaded from a pirate site. Maybe it was copied into a massive collection of books used to train an AI model. Maybe you never gave anyone permission to use it that way. And maybe you never even knew it happened! For thousands of authors, this isn’t some far-off possibility. It’s already happened.
In 2023, The Atlantic investigated a dataset called Books3 and found that it contained more than 191,000 books. The dataset was built from pirated ebooks, and at least 183,000 of the books could be connected to author information. Many of those authors had no idea their work was being used to train AI.
And Books3 is only one part of a much bigger story. As artificial intelligence becomes more powerful, books aren’t just being used as training material. They’re being summarized, turned into audio, transformed into videos, and repackaged in ways authors and publishers never agreed to.
So what exactly is happening to your book once AI gets its hands on it? Is your work being used somewhere you don’t know about? And what can you actually do about it?
Let’s take a look.
Why Are Books So Valuable to AI?
To understand why this is happening, it helps to understand how AI systems learn. They’re trained on enormous amounts of text. By processing all of that material, they learn patterns in language, how people write, how ideas are structured, how stories unfold, how questions are answered, and so on. Books happen to be really good material for this.
Think about it. A book is hundreds of pages of edited, organized, coherent writing. A novel has characters, dialogue, plot, and narrative structure. A nonfiction book might contain years of research and expertise. Compared with a random webpage or social media post, there’s a lot more substance to work with.
That makes books valuable training material. It also makes them valuable targets.
The Books3 investigation showed just how easy it was for copyrighted books to end up in these datasets. The collection came from pirated ebooks, meaning the authors didn’t have a say in whether their work was included.
Imagine spending five years writing a book, finally publishing it, and then discovering that a copy of it has been quietly passed around and used to help build an AI system.
How would you feel if that were your book? And more importantly, can companies actually do that?
What Does Copyright Law Say?
There’s no simple answer yet. Copyright law and AI are still working through a lot of very new territory. But one recent case made something very clear: piracy has consequences.
In Bartz v. Anthropic, a group of authors sued Anthropic over its use of books that had been downloaded from pirate sites and used in connection with training its AI models.
The case ended with a $1.5 billion settlement covering nearly half a million books. The Association of American Publishers called it the largest class action copyright settlement in history. The settlement also requires Anthropic to destroy the pirated books involved and copies derived from them.
That’s a huge number. And it’s a pretty significant moment for authors and publishers.
AAP President and CEO Maria A. Pallante put the industry’s position pretty clearly: “Partnerships, not piracy, are the best path forward.”
So does that mean the AI copyright debate is over? Not even close.
There are still plenty of questions around how copyrighted works can be used to train AI. But one thing is becoming harder to ignore: using pirated content as a shortcut to build AI systems can have very real consequences.
Your Book Can Be Used by AI in More Ways Than One
Here’s where things get even more interesting. Training is only the beginning.
Even if you put the question of AI training aside, generative AI has made it incredibly easy to turn an existing book into something new.
Take audiobooks. We’ve already seen AI generated versions of copyrighted books show up online without the author’s or publisher’s involvement. The technology makes it possible to create an audio version of a book without hiring a narrator, booking a studio, or going through the traditional publishing process.
Then there are AI generated book summaries. Want the main points of a 400 page book in five minutes? AI can do it. Want that summary turned into a video? That’s possible too!
And then there are tools like Google’s NotebookLM, which can take source material and turn it into a conversational, podcast style discussion. There’s an obvious benefit here. Information can suddenly become much easier to understand and consume. But here’s the tricky part: where is the line between making information more accessible and creating a new product from someone else’s work?
We’re starting to see similar questions around AI generated video. Stories can be transformed into visual content. Characters can be recreated. Pieces of existing works can be repackaged in ways that would have taken significant time and money just a few years ago.
What happens when that technology is used with your book? That’s the question authors and publishers are going to be asking more and more.
AI hasn’t just made copying easier. It has made transforming and redistributing content easier, too.
So What Should Authors and Publishers Do?
For a long time, protecting a book online meant finding the problem after it happened. Someone discovers an unauthorized copy. A publisher sends a takedown request. The copy comes down. Then another one appears somewhere else.
Sound familiar?
That model gets a lot harder when you’re dealing with AI. There can be thousands of copies and variations of content spread across different platforms. And by the time someone finds one manually, there may already be more.
So what if, instead of waiting for someone to find the problem, you could actively monitor for it? That’s where the industry is starting to shift.
In May 2026, the Association of American Publishers announced a partnership with Vermillio to help identify and remove infringing copies of literary works, including unauthorized generative AI versions of audiobooks. The AAP described the problem as a “rapidly rising flood” of AI generated copies that reproduce, mimic, and perform protected audiobooks.
The goal isn’t just to find infringement eventually. It’s to find it sooner and make it easier to do something about it. Vermillio CEO Dan Neely described the shift as “moving strategically from defense to offense in the AI era.” And that’s really where things are headed.
Instead of relying on authors and publishers to constantly search the internet themselves, tools like TraceID can help monitor a catalog continuously and flag unauthorized uses as they appear. TraceID’s Comprehensive Catalog Protection, Content Monitoring, and Automated Takedowns give rights holders a way to stay ahead of the problem rather than constantly playing catch up.
The Bottom Line
AI has learned from a huge amount of human created content. And for many authors, that happened without them ever agreeing to it. Books3 showed us how widespread the problem could be. Bartz v. Anthropic showed that using pirated books can carry a very real price tag. And now, we’re seeing AI create entirely new ways for books to be copied and transformed after publication.
Technology is moving incredibly fast. Copyright law is trying to catch up. And authors and publishers are left trying to figure out what all of this means for the work they’ve spent years creating.
So where does that leave you?
First, know what’s happening to your work. Where is your book showing up? Is someone using it to create something new? Is there an unauthorized audiobook online? Has your content been copied or transformed somewhere you don’t know about? Because you can’t protect what you can’t see.
The future of publishing and AI is still being written. But authors and publishers don’t have to sit back and wait to see what happens. In the AI era, finding out about infringement months later isn’t really protection. It’s just finding out what already happened.
