AI companies are buying old books—and feeding them into shredders
There is something particularly obscene about destroying an old book. It feels a lot like destroying an ancient statue.
This isn’t about throwing away a battered copy of a bestseller with another million copies floating around. Think of a book that has survived for decades—perhaps even a century—passing from one owner to another while wars were fought, governments fell and nearly everything around it disappeared.
When the Taliban blew up the ancient Buddhas of Bamiyan, photographs already existed. Their dimensions had been recorded. Their images could be reproduced.
But the statues were gone.
A digital copy is not the original object.
AI companies are now buying old books in bulk, cutting off their spines, scanning the loose pages through high-speed machines and shredding whatever remains—all so the machines can read them faster.
Apparently, they are particularly interested in books published before 2022 because they can be reasonably certain those books were written by human beings rather than generated by AI.
One company facilitating these purchases is ISBNdb, which offers to help AI companies acquire hundreds of thousands—or even a million—physical books at a time. The company understands how this looks. Its own website reportedly warns prospective clients that the “optics problem is real” because “‘AI company destroys two million books’ is not a headline that generates sympathy.”
No, it certainly isn’t.
ISBNdb offers confidentiality agreements, keeps buyers anonymous and suggests describing the process as “digital preservation.”
That’s a convenient euphemism, but they aren’t preserving the books.
They are destroying the books to train AI.
Anthropic did this on an enormous scale through a secretive effort called Project Panama. A hydraulic cutter removed each book’s spine. The pages were trimmed, separated and fed through industrial scanners. The digital copies went into Anthropic’s private library. What remained of the books was sent away for recycling.
Justine Bateman, the filmmaker and former Family Ties actress, called it what it essentially is: “burning books.”
.@ISBNdb_com @AnthropicAI et al, in addition to stealing your training material, you’re essentially “burning books.” Everybody who runs and works at these companies are legitimately disgusting individuals. https://t.co/EEZe0siylZ
— Justine Bateman (@JustineBateman) July 27, 2026
That sounds dramatic until you consider what is actually happening.
These companies are taking books written by other people, edited by other people, published by other people and preserved by generations of owners. They slice them apart because turning the pages carefully would take too long.
Consider the nerve of these people.
None of this is necessary. Books can be scanned without destroying them. Libraries, museums and archivists do it every day. It is slower and more expensive—which, of course, is the problem—but that is what responsible people do when they handle objects that may not be replaceable.
Elon Musk made that point after the story began spreading across X. He said he had instructed his SpaceXAI team to preserve any rare books in a library and “scan them the hard way” rather than cutting off their spines.
I’ve asked the SpaceXAI team to preserve any rare books in a library and scan them the hard way vs just cutting off the spine and scanning
— Elon Musk (@elonmusk) July 27, 2026
Good.
Because the physical book matters.
A digital scan can preserve its words and pictures, but it cannot preserve the object itself: the binding, the paper, the printing method, the handwritten notes, the ownership marks or the simple fact that this particular book passed through the world and survived.
Booksellers told 404 Media that rare and obscure books with very few surviving copies are being pulled into this system. When books are purchased by the hundreds of thousands, who is stopping to determine whether a particular edition is historically significant—or whether another copy exists at all?
Then comes the final insult.
After helping itself to the work of millions of authors, Anthropic has become deeply concerned about competitors using its output to train their models.
David Sacks summarized the arrangement perfectly:
Anthropic maintains that it is entitled to train for free on all the world’s output, even if the author objects. But if a competitor trains on Anthropic’s output after paying for it, that is IP theft. The hypocrisy is breathtaking. https://t.co/ZosLMzlpf7
— David Sacks (@DavidSacks) July 27, 2026
Anthropic believes it is entitled to train on the world’s output for free—even when the people who created it object. But when a competitor pays to access Anthropic’s output and trains on that, Anthropic calls it intellectual-property theft.
The rule is apparently very simple:
Human knowledge belongs to Anthropic.
Anthropic’s knowledge belongs to Anthropic.
They destroy the books, keep what was inside and somehow walk away believing that they are the owners.



