404 Media tracked a rare-book shipment to an Amazon AI training facility

404 Media physically tracked a shipment of used and rare books and found it ended inside an Amazon AI training facility, where the books are being cut apart and scanned to feed the company's AI models. The investigation, published by 404 Media, followed the parcel from a used-book seller to a warehouse operation whose purpose appears to be converting printed pages into training data.

What 404 Media found

Reporters placed a tracker on a box of secondhand books and watched it travel to an Amazon-linked site built for large-scale digitization. According to the report, workers there slice off the spines, a step known as guillotining, so pages can run through high-speed scanners. The physical book is destroyed in the process; what survives is the text, ready to be ingested by a model. TechCrunch summed up the irony bluntly: the company that started as an online bookstore is now shredding books to build AI.

A quiet buying spree

Used-book sellers have noticed something odd for months, sudden bulk orders that clear entire shelves at once. Gizmodo reports that many dealers are racking up sales they are not thrilled about, suspecting the buyers are AI companies hunting for training material. Fast Company describes firms acquiring used books by the thousands, a cheap way to source clean, edited, human-written prose that the open web cannot match. Printed books are attractive precisely because they are curated: professionally written, copyedited, and free of the spam and low-quality filler that pollutes web-scraped datasets.

The copyright and culture question

Buying a physical copy and scanning it sits in murkier legal territory than scraping pirated text, but it is far from settled. Publishers and authors are already fighting over whether training on books counts as fair use, and owning a paper copy does not automatically grant the right to reproduce its contents at scale. Destroying rare or out-of-print volumes raises a separate worry: some of these copies are scarce, and cutting them apart removes them from circulation for good. The prose becomes model weights; the object is gone.

The takeaway

The scramble for book-length, high-quality text shows how hungry modern systems are for data, the same pressure driving open-weight releases like Qwen and reshaping who controls the AI infrastructure layer. It also underlines why understanding agentic AI and using these tools well matters more each month, a theme covered in Techpresso's guide to AI for productivity. For now, one image is hard to shake: Amazon, once a bookseller, feeding rare books into a scanner one severed spine at a time.