The Books They Break to Build AI: The Hidden Cost of Artificial Intelligence
Our books, our art, our stories and our creativity are worth more than the data they can provide. As AI companies race to build ever-larger models, we need to ask an uncomfortable question: who is paying the price for this technological progress?
There is something wonderfully human about a book. Yet, in the race to build increasingly powerful artificial intelligence, creative work is increasingly being treated as something else entirely: raw material.
Across the technology industry, companies have assembled enormous collections of writing, images and other creative works to train their AI systems. Some of these collections have included copyrighted books obtained without their authors' permission. And in a particularly unsettling development, companies have bought physical books, cut off their bindings, scanned their pages and discarded the original volumes, essentially breaking book for the betterment of AI.
Yes, really.
This may seem like book-loving people's oversensationalisation of a hypothetical concern about what AI might do in the future. That is not the case, though. Court records and investigative journalism have documented specific instances of these practices. They raise difficult questions about copyright, consent, fair compensation and the kind of creative world we are building. For anyone who loves books, supports independent artists or believes that creativity deserves to be valued, this matters.
In This Article
Project Panama: When books become disposable
In January 2026, The Washington Post published an investigation into newly unsealed court documents revealing the extraordinary lengths to which AI company Anthropic went to obtain books for its AI systems.
The reporting described an internal initiative called Project Panama. In a planning document, the project was described as an effort to “destructively scan all the books in the world”.
The idea was as straightforward as it was disturbing: acquire vast quantities of physical books, remove their bindings, scan their pages into digital files and recycle the paper originals.
According to the reporting, Anthropic spent tens of millions of dollars acquiring millions of books, often in large batches. A scanning process involving industrial cutting machinery made it possible to process the pages quickly, turning physical books into machine-readable digital material. The project was intended to supply high-quality text for AI models, including those behind Claude, Anthropic's chatbot.
You can read the investigation here:
There is an important distinction to make here. Anthropic's physical book-scanning operation was not, by itself, the same legal issue as its acquisition of pirated digital books. A court found that using lawfully acquired books to train its AI models qualified as fair use in the circumstances before it, while the company's acquisition and retention of pirated copies raised separate copyright concerns. The latter claims were subsequently settled.
But the practical reality remains startling. A company seeking to build an AI system treated millions of books as material to be processed, extracted and discarded. Imagine spending years writing a novel. Imagine the years before that, spent learning your craft, studying the world, developing your voice and finding the courage to put your work into it. Imagine holding the finished book for the first time. Now imagine that, somewhere along the way, your work becomes one item in a huge acquisition programme. Its pages are cut apart, digitised and incorporated into a corporate resource. The physical object is recycled, while the information extracted from it may continue to be useful to the company long after the original book is gone.
Books are extraordinary things. People across generations and continents can all read the very same words––isn't it an incredible thing that you can read the thoughts of someone who lived hundreds of years before you? The physical form of a book has value, and so does the labour that created its contents. And yet, in this process, the book can become little more than a disposable vessel for data.
And it isn't just Anthropic
If you remember hearing about Amazon and books being dismantled for AI, you were remembering something real, too.
In August 2026, Ars Technica reported on an investigation by 404 Media that traced a book from a bulk order to an Amazon facility in Las Vegas. A tracking device placed in a book helped connect the shipment to a facility where books were being dismantled and scanned.
Amazon responded that it purchases books through commercial channels to help develop and improve its products and services. Its statement did not explicitly confirm that the books were being processed for AI training. The reporting connected the operation to Amazon's broader AI ambitions, but that distinction is important: the precise purpose of the particular scanning operation was not publicly confirmed in Amazon's statement.
Read the reporting here:
The report added to growing concerns among booksellers that AI companies are buying large quantities of books, including difficult-to-find titles, to obtain material that may not already exist in easily accessible digital collections.
Why would a technology company want so many books?
Because modern AI systems need enormous amounts of training material. Text helps language models learn statistical patterns in language: how words relate to one another, how sentences are structured, how arguments develop, how stories unfold and how different forms of writing work. The more varied and useful the training material, the more capable a model may become at certain tasks.
Books are particularly valuable in this context. They contain sustained arguments, carefully constructed narratives, specialised knowledge, distinctive voices and complex relationships between ideas. Unlike much of the short, repetitive text found online, a novel or a scholarly work can provide thousands of pages of developed language and thought. This helps explain why books have become such a sought-after resource in the AI industry.
It also makes the ethical question harder to ignore.
If human writing is valuable enough for companies to spend millions acquiring it, cutting it apart and converting it into training material, why are the people who created that writing so often left out of the conversation?
The other side of the story: books obtained without permission
Physical book destruction is shocking, but it is only one part of the wider controversy.
In Bartz v. Anthropic, a copyright case brought by authors Andrea Bartz, Charles Graeber and Kirk Wallace Johnson, a US federal judge examined how Anthropic had acquired and used copyrighted books. The court's June 2025 ruling described the company's acquisition of millions of digital books from pirate websites, alongside its purchase and scanning of physical books. The digital copies included books obtained from unauthorised online collections, including Books3 and Library Genesis, commonly known as LibGen.
The court found that Anthropic's use of books to train AI models could qualify as fair use. However, it distinguished that training use from the company's acquisition of pirated books and its retention of a permanent digital library. The fact that a company uses material for AI training does not automatically make every step in obtaining, copying or storing that material lawful.
The case was subsequently settled for $1.5 billion, without Anthropic admitting wrongdoing. The settlement addressed claims relating to the acquisition of copyrighted books, rather than establishing that every form of AI training on copyrighted material is unlawful.
The original court ruling is available here:
And here is the reporting on the settlement and the wider investigation:
Anthropic is not the only company facing questions about its training material. Authors have brought copyright lawsuits against companies including OpenAI, Microsoft and Meta, alleging unauthorised copying or use of their work. These cases involve different evidence, claims and legal questions; their outcomes should not be treated as interchangeable.
What does it mean to take someone's work?
When people talk about AI stealing from artists and authors, the word stealing can mean several different things. It can refer to the unauthorised copying of a copyrighted work. It can describe the use of creative material without consent or compensation. It can also express a broader ethical objection: that companies have built valuable products using the accumulated work of human creators without adequately recognising the people whose labour helped make those products possible.
These are related concerns, but they are not identical. Whether a particular use infringes copyright depends on the facts and the applicable law. Not every use of a copyrighted work for training has been found unlawful, and the law remains contested in many jurisdictions.
Nevertheless, the ethical problem is clear enough to discuss.
Authors do not produce books in a vacuum. They spend years learning, researching, drafting, revising and editing. Illustrators develop techniques over time. Painters build a visual language through practice. Designers learn to recognise the small decisions that make a piece of work feel balanced, expressive or original. All of this labour produces the creative material that AI developers want their systems to learn from. And when that material is copied at scale without meaningful permission, the creator can lose control over how their work is used, even while the company collecting it gains a valuable resource.
The imbalance is difficult to defend. A corporation can transform millions of works into an asset that supports a commercial product, while individual creators may not even know their work was included.
There is a difference between being inspired by art and systematically collecting vast quantities of art to build a competing commercial technology. There is a difference between learning from a book as a reader and copying millions of books into a corporate training library. And there is a difference between sharing in a culture of creativity and treating that culture as a free resource for private profit.
The distinction deserves to matter, even when the legal boundaries are still being worked out.
Artists are facing the same problem
This debate extends far beyond books. Illustrators, photographers, painters, graphic designers, concept artists and other visual creators have raised concerns about their work being collected and used to train image-generation systems without their consent. In many cases, artists say they were never asked whether their work could be used. They were not offered payment, and they were not given meaningful control over whether their images became part of a training dataset.
The resulting systems can generate images in response to written prompts, sometimes producing work that resembles particular artistic styles or visual traditions. That does not mean every generated image is a direct copy of a specific artwork. The technical process is more complicated than retrieving and pasting individual images, and the legal status of training and outputs depends on the circumstances.
But the concerns are not imaginary. An illustrator may spend years developing a recognisable style, building a portfolio and finding customers who appreciate their work. A client can then use an image generator to request something in a similar aesthetic, potentially reducing the need to commission the artist at all.
This is not saying that artists should be immune to new tools or that artistic influence must stop. Art has always developed through conversation, reference, experimentation and reinterpretation. The issue, rather, is the scale and power of the systems involved, and whether creators have a meaningful say in the use of their work. Human artists are not simply competing with other people who have spent years practising their craft. They may be competing with commercial products trained on vast collections of creative work, including work made by people who never agreed to participate. The people whose work helped make those products useful may then find themselves competing against them in the marketplace.
That is a deeply uncomfortable arrangement. And it is one reason the debate about AI should include not only technological progress, but also the livelihoods, rights and dignity of the people whose creativity makes that progress possible.
The hidden cost of making creativity cheap
One of the most troubling aspects of this debate is the way it intersects with the economics of creative work.
Most authors and artists are not wealthy. Many balance creative work with other jobs. They deal with uncertain income, rising costs, rejection, limited opportunities and the constant challenge of persuading people that their work is worth paying for. A book may take years to write but generate only modest royalties. An illustration may take days to complete, even after a lifetime of practice. A small independent artist may rely on a handful of commissions or product sales to make their work sustainable. For these people, the ability to earn a living from creativity is not an abstract philosophical question. It determines whether they can afford to keep creating.
AI companies, meanwhile, are building commercial products that depend on processing extraordinary quantities of information. The technology can be sold through subscriptions, business contracts and other revenue-generating services.
When the creative material supporting these products has been obtained without permission or payment, the benefits and burdens are distributed unevenly. Creators supply the work. Technology companies build the systems. The resulting products may compete with the very people whose labour helped make them possible.
This does not mean every AI product necessarily harms every artist or author. Some creators use AI tools in their own workflows, and some may benefit from new opportunities these technologies create. Nor can we assume that every AI-generated work replaces a human commission. But it is reasonable to ask whether a business model built partly on uncompensated creative labour is sustainable or fair.
If human creativity is valuable enough to build an industry around, the people who create it deserve a meaningful place in that industry's economy.
Why classic novels and old books matter, too
You might reasonably wonder whether this matters if a book is old, out of print or considered a classic.
After all, many famous novels are in the public domain, meaning their copyright protection has expired or does not apply. Works such as Jane Austen's Pride and Prejudice and Mary Shelley's Frankenstein can often be read, reproduced and adapted without obtaining permission from the original author or their estate, subject to the relevant jurisdiction and the particular edition being used.
But public domain does not mean that every book is free of copyright restrictions. A modern translation of an old novel may still be protected. So may a new introduction, critical commentary, illustration or specially prepared edition. A contemporary novelist's work is generally protected even if a technology company would find it useful to train a language model on it.
There is also a distinction between legal ownership and cultural value. Even where a text can legally be copied, we can still ask what it means to preserve and value literature. Is a book merely a sequence of words that can be extracted and processed, or is it part of a living tradition of readers, writers, publishers, booksellers, translators, editors, illustrators and collectors? Classic literature survives because people continue to read it, teach it, discuss it, publish it and pass it on. Independent bookshops, libraries, publishers and literary communities help keep these works accessible and meaningful.
The physical book matters, too. Its typography, illustrations, annotations, cover design and paper can tell their own story. A well-loved paperback with a cracked spine or a second-hand book bearing a previous reader's handwritten note can carry a history that no training dataset captures. There is something profoundly ironic about a technology industry extracting knowledge from books while potentially contributing to a culture in which the human labour and physical objects behind that knowledge are treated as expendable.
What would ethical AI actually look like?
Criticising current practices does not require us to reject every application of artificial intelligence. Nor does it require us to pretend that the technology will disappear if we refuse to use it. Instead, we can insist that innovation comes with responsibilities.
An ethical approach to AI and creative work should begin with several basic principles.
Consent should be meaningful. Creators should have clear information about whether their work is being used to train commercial AI systems and meaningful ways to decide whether to participate, where appropriate. A company should not assume that material being accessible online automatically means its creator agrees to every possible use.
Permission should be respected. Where a licence is required, companies should obtain it. Copyright law varies between countries, and some uses may be lawful without individual permission, but legal uncertainty should not become an excuse to dismiss the interests of creators.
Compensation should be fair. When businesses derive substantial commercial value from creative work, licensing and collective remuneration deserve serious consideration. Individual artists should not have to fight enormous corporations one by one simply to have their interests acknowledged.
Transparency should be the norm. Companies should explain, as clearly as practicable, what kinds of material they use to train their systems, how that material is acquired and what controls are available to rights holders.
Human creativity should remain visible and valued. A generated image or passage of text is not automatically equivalent to a work made by a human being. The time, skill, intention, experience and personal perspective behind human-made art deserve recognition, even (or especially so) in a world where automated tools are increasingly available.
These principles are not unreasonable demands for a "perfect" world. We can build them as foundations of a healthier relationship between technology and the people whose work it draws upon.
Organisations such as the Authors Guild have been campaigning on AI and authors' rights, while artists and other creative communities continue to push for stronger protections and more transparent practices.
The debate is ongoing, and no single policy will resolve every difficulty. But progress should be measured not only by what machines can produce, but also by whether the people who make our culture are treated fairly.
What can we do as readers, artists and consumers?
It is easy to feel powerless when confronted with enormous technology companies and legal disputes involving millions of books. But we are not entirely powerless.
We can support independent bookshops, small publishers and authors whose work we love. We can buy directly from artists when possible, commission human illustrators, purchase books from independent creators and choose products made with care rather than mass-produced imitations.
We can ask companies how their AI products are trained and whether they respect creators' rights. We can support organisations advocating for fair licensing and better protections. We can pay attention to the policies of the platforms and services we use.
And we can think critically about the convenience of generating something instantly. If we want a particular illustration, perhaps we can commission an artist whose style we admire. If we want a book recommendation, we can ask a bookseller, join a reading group or explore reviews written by people who genuinely love literature. If we need a gift for someone who adores books, we can choose a thoughtfully designed bookmark, an artist-made print, a reading journal or a carefully selected edition of a favourite novel.
We need to remember that our spending, attention and habits help shape the world around us. Every time we pay a human creator for their work, we help make room for them to create more. Every time we recommend an independent artist, share an author's work or choose a human-made product, we help build a culture that values creativity as something more than a source of data. And every time we ask questions about the origins of a technology, we make it harder for businesses to treat those origins as irrelevant.
A final thought: creativity is not just data
This topic is complicated, and will likely only grow more complicated over time, as AI models get better and better at imitating humans, to the point we will not be able to tell what is human-made and what was created by AI. People use AI for many different reasons, with different levels of knowledge, access and choice. We should make room for honest discussion rather than assuming that everyone who has ever used an AI tool is indifferent to human creativity.
It would be silly and a very, very difficult task to ask everyone to achieve personal purity in a world where these technologies are increasingly embedded in everyday life. Instead, what we can do is make more informed decisions, demand better practices and refuse to accept that technological convenience must always come before human interests. Because books are not just training material, and art is not just a collection of patterns AI can piece together.
A writer's years of work are not merely a resource to be extracted. An illustrator's style is not simply a feature to be replicated. And a human being's creative contribution does not become less meaningful because a machine can produce something that resembles it.
Creativity is one of the ways we make sense of being alive. It allows us to tell stories, preserve memories, challenge injustice, imagine different futures and recognise one another across differences in time and place. It deserves to be treated with care.
The next time you buy a book, discover a new illustrator, choose a gift or commission a piece of art, consider the person behind it. Look for the independent creators whose work makes your world more interesting. Support the businesses that put human creativity at the heart of what they do.
And when you use AI, take a moment to ask yourself what you are using it for, whose work may have contributed to it and whether there is a more ethical way to achieve your goal.
We do not have to reject every new technology to demand that it respects the people who make our culture. We do, however, have to be willing to ask difficult questions, even when the answers are inconvenient. And let's start thinking more carefully about what ethical AI use really means. Because a world full of instantly generated content is not necessarily a world richer in creativity.
We're giving classics and new releases a home. First and foremost an online book- and art store. Committed to cataloguing real, human stories & art in the age of the artificial, and to leave our ecological footprint as small as possible. All of our products are created or printed only when an order is placed, so no item ends up in a landfill.
Our collection spans across multiple mediums, including novels, art works (and prints), poetry collections, bookish items like stickers & clothing, photography works, home decor, meditation, zines and more.
Work with us
Interested in working with us? Click the links below to find more information. If you want to pitch a zine or partnership, please reach out through an email.
Contact Us
For inquiries, collaboration opportunities, or any assistance, please don't hesitate to get in touch:
Email: info@apolloimperium.com
Comments