If you’ve ever sold an old or rare book online, there’s a chance it didn’t end up on someone’s shelf. Instead, it may have been shipped to a warehouse in Las Vegas where the book’s spine is removed and the pages are fed through a scanner so the text can be digitized for use in training artificial intelligence systems.
A recent investigation from 404 Media documented one such shipment. Working with a bookseller who received an unusual order for 1,000 books through Biblio, an online marketplace that allows buyers to remain anonymous, investigators hid an Apple AirTag inside a book and tracked its journey. Weeks later, the device led them to an Amazon facility identified as VGT3.
An AirTag Blew the Whistle
Employees at VGT3 describe their primary task as scanning books. According to posts on internal forums and accounts collected during the investigation, some staff physically remove the spines of books so pages can be fed through high-speed scanners. Others scan barcodes and handle the logistics of processing large volumes of printed material. The result, several workers reported, is that the original physical book is destroyed and does not survive the conversion into digital text.
A distinctive logo in the part of the warehouse where this work happens — a dinosaur clutching a book — has been noted by those familiar with the facility. The logo, observers say, seems to reflect the literal dismantling of printed materials in order to extract their words.
Amazon provided a brief statement in response to the reporting, saying the company purchases books through normal commercial channels to help improve its products. The statement did not address why original copies are destroyed rather than kept, resold, or donated after digitization.
This kind of operation is not unique to Amazon. In a separate case reported last year, a lawsuit revealed that Anthropic, the company behind the Claude chatbot, had cut apart millions of printed books to scan them for training data. A judge ultimately ruled that transforming a physical book into digital text and destroying the original could fall under the legal definition of a “transformative” use and therefore did not infringe copyright under existing interpretations.
Why Rare Books Are the Target
Rare and out-of-print books are particularly attractive to companies building or refining AI models for a simple reason: their text is often not available anywhere online. That makes these works fresh sources of training material, providing language and perspectives that haven’t been replicated or diluted by prior AI outputs. Because many of these works were written long before the era of AI-generated writing, they can add unique phrasing, historical context, and nuance that are otherwise scarce in contemporary datasets.
Booksellers who have encountered these orders say the companies sourcing such material often treat books purely as repositories of words. They don’t account for the historical, cultural, or sentimental value of a rare or unique physical copy. For collectors, libraries, and donors, that human and archival value can be immense — yet it may be ignored when the primary objective is to extract text for machine learning.
There is currently no public indication that Amazon’s practices, as described in the investigation, violate existing law. Courts have so far tended to side with AI developers in disputes over similar book-scanning practices, viewing some forms of digitization and reuse as legally transformative. Nevertheless, the process raises important ethical and practical questions about stewardship and the fate of physical books once they enter third-party marketplaces or commercial supply chains.
If you’re considering selling or donating an old, rare, or out-of-print book through an online marketplace, it’s worth being aware that the copy might not reach another reader in its original form. Instead, it may be dismantled and incorporated into datasets that help train AI systems — a practical outcome that has consequences for preservation, access, and the long-term survival of printed works.