Tracked by an AirTag: Inside Amazon’s Secret Pipeline That Reportedly Destroys Physical Books

Tucked inside a bulk order of roughly 1,000 books purchased through the Biblio marketplace was something the buyer didn’t expect: an Apple AirTag. 404 Media placed it there. The book traveled from a bookseller’s hands through a processing warehouse and arrived at an Amazon facility in Las Vegas — where, according to workers, it was never coming back intact. Amazon is reportedly buying physical books, scanning them for AI training, and destroying the originals. This is documented investigative reporting, not speculation — and what it reveals extends well beyond one warehouse.

The Pipeline Nobody Advertised

Anonymous bulk orders, sliced bindings, and a dinosaur logo that says more than Amazon’s PR team intended.

Amazon’s Las Vegas team, reportedly called VGT3 — whose logo depicts a dinosaur clutching a book in its teeth — operates a straightforward workflow: receive massive shipments of printed books, slice off the bindings, scan the loose pages, discard what’s left. (The dinosaur detail is darkly funny until you remember what happened to the things dinosaurs ate.) When asked, Amazon issued a statement: “Amazon purchases books through commercial channels to help develop and improve the products and services our customers use.” What the statement doesn’t address: what happens to the books afterward.

The mechanics are worth understanding. Bulk orders are placed anonymously through marketplaces like Biblio, routed through intermediaries, and delivered to facilities where workers describe destructive scanning as standard workflow. Anthropic’s internal “Project Panama” documents, revealed in 2026 litigation, show a near-identical process at scale — millions of books bought, spines sliced, pages scanned, originals discarded.

Here’s why printed books specifically. Pre-2022 volumes are very likely free of AI-generated text — no machine output contaminating the corpus, dramatically reducing the risk of model collapse from systems training on their own synthetic output. ISBNdb, a book-metadata company, briefly marketed this logic directly: “The world’s best AI training data is sitting on a shelf.” They pulled the page after 404 Media reported on it, calling it “a test of market interest” that was never operational.

“The world’s best AI training data is sitting on a shelf.” — ISBNdb marketing copy, since removed

This isn’t just Amazon. Booksellers across Europe received bulk orders for thousands of titles — some initially dismissed them as phishing attempts before realizing the orders were likely tied to AI training procurement. One intermediary reportedly acknowledged, according to Boing Boing, that “‘AI company destroys two million books’ is not a headline that generates sympathy.” Which is exactly why these operations run under NDAs and anonymous marketplace accounts, well out of public view.

“‘AI company destroys two million books’ is not a headline that generates sympathy.”
— Source quoted in Boing Boing reporting

Legal, Quiet, and Largely Unregulated

A 2025 court ruling effectively cleared the legal path — and no regulatory body has stepped in since.

In June 2025, U.S. federal Judge William Alsup ruled that scanning legally purchased books for AI training can constitute transformative fair use — specifically covering lawfully acquired print copies converted to digital for model training. That ruling legitimizes the core pipeline, at least for legally obtained books. No regulatory body currently monitors how many books are bought, scanned, or destroyed for AI data centers and training purposes. Authors receive the original sale price. Nothing more.

Your local used bookshop and the rare-title dealers serving serious collectors are now quietly competing with AI labs that have no price sensitivity and no interest in keeping the books. The physical artifact of human knowledge — centuries of edited, organized thought — is becoming feedstock. That’s not a metaphor. That’s what the AirTag found.