You think your old books are safe on the shelf? Think again. Large language model creators are quietly buying up massive quantities of secondhand books, shipping them to industrial hubs, and running them through high-speed scanners. The catch? To scan the pages fast enough, they chop off the spines with industrial guillotines. The physical volumes are completely destroyed in the process.
This isn't a villainous plot crafted by comic book masterminds. It's corporate apathy. Tech companies aren't trying to burn history out of hatred for literature. They just want clean, high-quality human prose, and they don't care what gets broken along the way.
Following the Paper Trail to the Shredder
The whole operation came to light when investigative reporting by 404 Media tracked a suspicious bulk book order. A used-book seller noticed an anonymous buyer scooping up hundreds of titles at once through platforms like Biblio. Curious about where the inventory was going, the seller hid an Apple AirTag inside one of the volumes.
The tracker led straight to an Amazon warehouse operation in Nevada, specifically a site designated as VGT3. Workers at the facility revealed a mundane yet jarring reality. Their daily routine involved stripping spines off thousands of books so the loose pages could feed industrial document scanners at maximum velocity.
Why destroy them? It's cheaper and faster. More importantly, court rulings regarding copyright and format-shifting have created a specific legal loophole. By purchasing a physical copy lawfully, scanning it, and destroying the original so no duplicate copy circulates outside internal systems, AI firms try to claim "fair use" protection under legal precedent like the Bartz v. Anthropic ruling.
The Myth of the Lost Treasure
Whenever the public hears that "rare books" are being fed into an AI woodchipper, panic ensues. People imagine first editions of classic literature or irreplaceable historical manuscripts going under the blade.
Let's clear that up. Most of these texts aren't lost literary masterpieces. They are niche manuals, old software guides, obscure textbooks, and out-of-print reference books that nobody read when they were published. They are rare in the sense that only a small number of copies exist, mostly because demand was zero.
Even so, the process raises profound cultural questions. When every surviving copy of an obscure 1995 technical manual gets its spine sliced off to feed an algorithm's training corpus, that physical artifact is gone forever. The text lives on as vector embeddings and model weights, but the object disappears.
The Search for Clean Data
Why go through the logistical nightmare of buying physical books when the internet is full of text? Simple. The open web is choked with spam, machine-generated noise, and low-quality data poisoning. AI models trained entirely on modern internet junk start to degrade.
To build smarter models, labs need professionally edited, human-written sentences. Printed books offer pristine grammar and structured prose that web scrapers can't replicate. That's why supply chains have emerged to quietly funnel physical books from anonymous marketplaces straight to automated scanning facilities.
You're left with an absurd irony. Amazon started as an online bookstore designed to ship physical books to readers everywhere. Now, parts of its vast logistics empire are dedicated to tearing books apart so machines can learn how to talk.
Nobody loses sleep over a destroyed software manual from thirty years ago. But the total indifference toward the physical history of print shows just how little weight human culture carries when weighed against computational scale.
Check your shelves. If you have old reference books you don't want reduced to pulp, keep them close.
Thousands, if not millions, of books sent to AI woodchipper
This report provides a detailed breakdown of the investigation into physical books being acquired and destroyed to feed artificial intelligence training data.
