For over a year, booksellers have suspected that AI companies were buying up rare books by the thousands and destroying them after scanning. They had no proof — until now. 404 Media tracked a shipment of rare books using a hidden AirTag and discovered it ended up at an Amazon AI training facility in Las Vegas, where a team systematically rips books from their spines and scans every page.
The facility, internally called VGT3, even has a logo on the door: a Tyrannosaurus rex preparing to devour a book.
How the Investigation Worked
404 Media connected with a bookseller who agreed to plant an Apple AirTag inside a rare book that was part of a bulk order. The AirTag was then tracked to the Amazon facility in Las Vegas, which houses a team focused on tearing books from their spines and scanning pages for AI training data.
Amazon declined to comment on the findings, only providing a generic statement: "Amazon purchases books through commercial channels to help develop and improve the products and services our customers use." Notably, the statement doesn't mention AI training at all.
Why Rare Books?
Amazon is developing what it considers frontier AI models — models that require massive amounts of unique training data to compete with Google, OpenAI, and Anthropic. Training on text from rare books that are difficult to find elsewhere offers a competitive advantage. Rivals like Anthropic and xAI have publicly stated they are not training on rare or antique books.
404 Media's investigation revealed that workers at the VGT3 facility were trained to scan barcodes or ISBNs before scanning books. This gives credence to booksellers' theory that AI companies are methodically working through the list of ISBNs to scan every printed book in the world, ensuring the highest volume of unique works in their training datasets.
The Supply Shortage
The investigation also uncovered that Amazon apparently ran low on books to scan earlier this year. Workers in online forums worried the warehouse might shut down if Amazon couldn't find more books. At one point, the supply completely ran out. But the facility is still operational, and bulk orders of rare books continue to be delivered and systematically destroyed.
The Ethical Dilemma
For booksellers, the money may be good, but the practice raises a serious ethical dilemma. Booksellers know how to assess the value of rare books — not just monetary value, but historical, intellectual, and sentimental value. AI firms seem to be skipping that assessment entirely, hunting low-cost, unique ISBNs to complete their checklists.
The books being targeted aren't necessarily prized first editions. They're often older books with lower monetary value — books never translated from obscure languages, books that were never widely distributed. But these works may still have historical and intellectual significance that AI firms overlook.
As the bookseller who planted the AirTag told 404 Media: a rare book's value can be derived from "all sorts of things" that "the AI companies don't care about. They just want the content as a bunch of words strung together."
The Broader Debate
The discovery has sparked debate about whether destroying rare books to train AI is inherently problematic. Some argue that many of these books have been sitting on shelves for decades gathering dust. Others counter that the real problem is the loss of knowledge — AI firms will "never share the contents" of books they scan, as they don't want anyone else to be able to train their AI on the same works.
Scottish bookseller Derek Walker told the BBC that AI firms should distinguish between works that won't be missed much and lesser-known antique works that may be "the only known surviving example of an edition." The loss of such a book to AI training would be irreversible.
"Even if you love the technology, you can admit that the concept of an AI literally eating books to become more powerful is pretty dystopian," one commenter observed.
What This Means
This investigation confirms what booksellers have suspected for a year: at least one major tech company is systematically destroying rare books to gain a competitive edge in AI development. The practice is legal — Amazon is purchasing the books through commercial channels. But it highlights the hidden costs of the AI race: not just electricity and water, but cultural heritage.
The books being scanned and destroyed will never be available to researchers, historians, or the public — only to the AI models that consumed them. And the companies doing the scanning have every incentive to keep it that way, because their training data is their competitive moat.
As AI companies guard their training data and use services that mask their identities as buyers, there's little desire to discuss bulk buying with booksellers. Allowing booksellers to weigh in would require a level of transparency that seems riskier than a possible reputation hit.
But now, at least, the practice is out in the open. What happens next depends on whether public pressure can force companies to adopt less destructive methods — or whether the race for training data will continue to consume the physical record of human knowledge, one ISBN at a time.
Top comments (0)