An AirTag found in a rare book pointed to an Amazon book scanning facility located in Las Vegas.

      A bookseller concealed an AirTag within a rare book and included it in a large shipment. The tracker traveled across the U.S. and ultimately arrived at an Amazon warehouse in Las Vegas. Emanuel Maiberg reported on this development for 404 Media on Monday. For a year, booksellers have suspected that AI companies were purchasing their inventory in bulk and then pulping it, but no one could identify a specific company until now. The existence of the Amazon book scanning operation had not been previously disclosed.

      Regarding Amazon's response, the company chose not to comment on the findings. It provided the same statement to 404 Media and Ars Technica, without mentioning artificial intelligence. “Amazon sources books through commercial channels to enhance and improve the products and services used by our customers,” the company stated. While this statement does not contradict the reporting, it also does not confirm it.

      At the warehouse, 404 Media discovered details about Amazon's book scanning team, known as VGT3. Employees explained that large shipments of printed books are received, the bindings are cut to enable faster page scanning, and the books do not survive the process. The logo on the door features a dinosaur grasping a book with its teeth bared, which Ashley Belanger described for Ars Technica as a Tyrannosaurus rex poised to consume one.

      404 Media also found discussions among workers at the facility. They mentioned that earlier this year, Amazon faced a shortage of books to scan, leading to concerns that the warehouse might close. At one point, there was an absolute lack of supply, but it remains operational.

      One significant finding supports a broader theory: booksellers had already speculated that buyers are targeting ISBNs instead of specific titles, as an ISBN designates a particular edition. Accumulating enough ISBNs verifies coverage of every printed book that has one. Upon reviewing the discussions among the workers, 404 Media discovered that staff scan the barcode or ISBN before scanning the book, lending additional credibility to this theory.

      This finding also clarifies the purchasing patterns that had puzzled booksellers; buyers often bypass celebrated first editions in favor of older books of minimal financial value, those never translated from a language few read today, or those that were never commercially significant due to limited printing.

      The bookseller who planted the tracker highlighted the implications of these actions for rare books, noting that their value is derived from historical, intellectual, and sentimental significance—factors that, in their view, AI companies disregard, only valuing the text as a sequence of words.

      The high demand for pre-2022 print books stems from a straightforward reason: anything published post-2022 may include machine-generated text, while models trained on AI content become less effective over time. Mike Pearl pointed out for Gizmodo that, prior to about 2022, books were exclusively written by humans. The supply of human-authored works from before the era of chatbots is finite and diminishing. This issue is not unique to English; we recently reported that China faced the same problem with Chinese-language materials.

      An additional comparison with Amazon's competitors is particularly telling. Anthropic informed Philippa Wain at the BBC that none of its data acquisition programs involve purchasing and destroying rare or antiquarian books. Ars Technica indicated that xAI has made similar statements publicly.

      Although Anthropic does engage in destructive scanning, court documents released last month detailed an initiative named Project Panama, aimed at destructively scanning all the world's books. The company paid $1.5 billion to resolve a separate copyright infringement claim related to pirated copies. Anthropic distinguishes between regular second-hand copies and rare or antiquarian works, while Amazon has not publicly set any distinctions.

      Amazon also utilizes material from its own platforms, as we reported last week that Twitch will provide data for training Amazon's AI unless users change their settings.

      The legality of destructive scanning is rooted in specific legal interpretations. Judge William Alsup ruled in 2025 that Anthropic's use of acquired books constituted transformative use, rendering it fair under copyright law. While purchasing a book and scanning it is lawful, the destruction of the original copy ensures one version exists before the scan and another afterward.

      According to Emily Hudson, an intellectual property expert at Oxford, the legal stance is different in the UK, where creating a training library and conducting training both necessitate the copyright holder's permission.

      The trade among booksellers is genuinely divided, and acknowledging this division is essential. Stuart Manley of Barter Books in Northumberland mentioned that he’s never encountered orders like this in three decades. Recently, one bulk purchase was equivalent to his typical weekly sales. He does not view much of it as a loss. “The world no longer requires five million copies of The Da Vinci Code,” he remarked. He noted that books that had remained unsold online for 20 years are now moving.

      Derek Walker of McNaughtan’s in Edinburgh remarked on the significance of certain titles, stating that losing an academic text printed in a