A bookseller hid an AirTag inside a rare book, let it go with a bulk order, and tracked it across the United States to a warehouse in Las Vegas owned by Amazon [1]. Booksellers had spent about a year suspecting that AI companies were buying their stock in bulk and destroying it after scanning, without being able to name one [2], so the change here is that a procurement rumour now has a party attached to it.
Emanuel Maiberg reported the finding for 404 Media on Monday [3]. Amazon calls the book scanning team at the site VGT3 [4]. Employees told 404 Media that large shipments of printed books arrive, staff slice the bindings off so the pages scan faster, and the book does not survive the process [5]. The logo on the door is a dinosaur holding a book and baring its teeth [6], which Ashley Belanger, writing the investigation up for Ars Technica, read as a Tyrannosaurus rex about to eat one [7].
Amazon declined to comment on the findings and gave Ars Technica the same statement it gave 404 Media, which does not mention AI: "Amazon purchases books through commercial channels to help develop and improve the products and services our customers use" [8][9].
The detail that does the most work is the barcode. 404 Media reviewed worker discussions and found that staff scan the barcode or ISBN before they scan the book [10]. That fits a bookseller theory that buyers are chasing ISBNs rather than titles, on the logic that an ISBN marks a distinct edition and enough of them covers every printed book carrying one [11]. It also explains a buying pattern sellers found baffling: celebrated first editions get skipped, while older books of low monetary value, never translated out of a little-read language, or never printed in commercially meaningful numbers, get bought [12]. What is being purchased is edition coverage, not literary value. The bookseller who planted the tracker told 404 Media the buyers want the content as a bunch of words strung together [13].
The premium on old paper is mechanical. Anything published since roughly 2022 may contain machine-written text, and models trained on model output degrade [14], a failure mode TechCrunch describes as model collapse [15]. Ars Technica notes Amazon is building what it considers frontier models, which need large volumes of unique training data, and that firms guard their data sources to protect an edge [16].
That is where rivals' positioning starts to matter to rights holders. Anthropic told the BBC that none of its data acquisition programmes buy and destroy rare or antiquarian books, and Ars Technica reports xAI has said the same publicly [17][18]. Anthropic does scan destructively: court documents unsealed last month named the effort Project Panama, with the stated aim of destructively scanning all the books in the world, and the company paid $1.5bn to settle a separate claim over pirated copies [19]. Amazon has drawn no comparable line in public [20].
The lawful basis is a reading of fair use. Judge William Alsup ruled in 2025 that Anthropic's use of purchased books was exceedingly transformative, and therefore fair use [21]. Emily Hudson, an intellectual property specialist at Oxford, told the BBC that the position in Britain differs [22].
Two things to watch. Worker forum posts reviewed by 404 Media said Amazon ran short of books to scan earlier this year, that supply ran out entirely at one point, and that some workers feared the site would close; it is still operating [23]. That is a demand signal, and it sits alongside The Next Web's report that Google paid $10mn for a bankrupt airline's internal records [24]. Also worth noting: all three published accounts of the Amazon finding trace back to the same 404 Media investigation [25], so the named-party fact currently rests on one outlet's work and one statement that confirms nothing.
