The motives are anti-human. The methods are draconian. And the end result is hard to distinguish from intellectual property theft at industrial scale. As Peter Biles noted here earlier, In an effort to skirt copyright and licensing headaches, AI labs are buying up millions of old physical books and destructively scanning their contents as training data for their LLMs (Large Language Models).
Where to begin discussing the problems with this practice? Let’s start by remembering that this scheme is the brain child of technicians, people who lead tech companies and tech labs who are wholly committed to a technological view of the world. Also known as techno-utopianism, it’s the belief that every single problem humans face can be addressed with technology, including the problems specifically caused by technology. There is nothing that can’t be enhanced, improved, ameliorated, or fixed with a technological invention or innovation. For people with this mindset, nothing is beyond the reach of technology. Nothing is sacred. Nothing is excluded.
Industrial-Scale Book Harvesting
The next thing to know is that this isn’t a handful of techie geeks walking into their local used bookstores and walking out with a handful of old books in their hands. This is industrial-scale book harvesting at its most audacious. Booksellers around the world are getting orders for hundreds or thousands of books at a time. Companies facilitating bulk acquisition of physical books have sprung up in recent years. Some try to downplay or hide their marketing efforts toward AI labs engaged in this practice. One such company, ISBNdb, recently boasted about its services on a landing page designed to attract business from AI clients:
Millions of the most valuable books have never been digitized. They exist only in physical form, scattered across library shelves, used bookstores, and out-of-print catalogues. We get them to you at scale.
Indeed, the ISBNdb website noted that up to a million physical books could be purchased in a single order. “The world’s best AI training data is sitting on a shelf,” claimed the website. After backlash, ISBNdb removed pages on their website that detailed these services to AI companies. But even if one such company has chosen to “pivot away from that direction,” other avenues remain for AI labs to acquire physical books at scale for such morally and legally dubious purposes.
From Digital to Physical Trawling
The next thing to remember is that AI and Big Tech companies can’t be trusted because they’re not driven by human-centric values and interests. Case in point: Project Panama. Before pivoting to legal methods of buying books, it was revealed that between 2021 and 2023, Anthropic co-founder Ben Mann had personally downloaded millions of pirated books online for use in training his company’s chatbot Claude.
In the face of serious legal problems stemming from this, the company quietly launched what they internally called Project Panama, a scheme that saw the company spend tens of millions of dollars acquiring, scanning, and then destroying physical books for AI model training. Anthropic chose an executive from Google Books to head up the project, and internal memos cautioned employees that Anthropic didn’t want the public to know what it was doing.
But despite the underhanded way Anthropic has gone about gobbling up “all the books in the world,” they still chalked up a major win in one of the many class action lawsuits that have been brought against them. In 2025, a federal judge ruled that Anthropic could use books to train its AI models under fair use law as long as they were purchased and not pirated. The case concluded this year with a $1.5 billion settlement to book authors and publishers, the largest ever monetary penalty in a copyright case. But $1.5 billion is chump change for a tech company like Anthropic. In reality, no monetary penalty is going to stop a Big Tech company from acquiring what it needs to win in the market.
Legal loopholes and court fines notwithstanding, is it ethical for AI companies to harvest massive amounts of copyrighted human wisdom and insight and use that source material to amalgamate proprietary responses it charges users to access? I would argue that it is not ethical and should not be tolerated.
So What Can We Do About It?
We can start by acquiring more old books ourselves! Before AI companies buy up all remaining physical copies of older, rare, out-of-print, obscure, or nearly forgotten books, we can buy our own copies. We’re at the point now where we can’t trust that editions of books we purchase on Amazon or digital books we get on Kindle are going to be unmodified, original, and true to the author’s original work. Acquiring older editions of books is a great way to guarantee authenticity. Yes, physical books take up space, but a generous library of books is a handsome sight indeed. Don’t let space stop you from having a good quantity of physical books on hand in your home.
Another thing we can do is distance ourselves from Big Tech companies that don’t have your interests at heart. I’ll say it again: we cannot trust AI and Big Tech companies to make honorable decisions based on human-centric values and interests. We must make those decisions, and if that means not using chatbots like Claude or ChatGPT, so be it. Take a stand for yourself and your family to acquire knowledge and wisdom the old-fashioned way: one book and lesson at a time, without shortcuts or distracting technological flimflam.
Cross-posted from The Human Adventure.