They put a tracking device in a shipment of rare books to see where it would end up. It ended up at an Amazon warehouse, where employees cut the binding off and fed it into a scanner.
The 404 Media investigation, published this week, traced the shipment to a Las Vegas facility called VGT3.
Workers there told the outlet their job is receiving massive shipments of printed books, scanning them, and destroying the physical copies as part of the process. Cutting the binding off a book is the fastest way to feed its pages through a scanner at volume, but it also makes the original unusable and unrecoverable once the process is done.
How the Investigation Actually Found This
The team at VGT3 has an internal logo: a dinosaur baring its teeth, holding a book.
Amazon didn't respond to 404 Media's request for comment on the report. The investigation marks the first public documentation of Amazon specifically running this kind of book-buying and destruction operation for AI training purposes.
Advocacy Groups Want the FTC to Step In
The 404 Media report landed days before more than a dozen public interest and consumer groups, including the Demand Progress Education Fund and the Consumer Federation of America, sent a letter to the FTC asking it to investigate the broader practice across the AI industry.
The letter argues AI companies are buying books in bulk, digitizing them, and destroying originals that in some cases are among the last surviving copies of a given work.
That's a distinct legal theory from a copyright claim. The FTC letter is arguing under Section 5 of the FTC Act, the provision covering unfair methods of competition, not under copyright law.
"AI companies are engineering a future where only the wealthiest incumbents can build high-quality AI models," the letter states, describing the practice as a potential violation of the FTC Act's ban on unfair competition.
The groups want the FTC to determine how widespread the practice is and how many destroyed books were irreplaceable.
Non-Destructive Scanning Already Exists
Destroying the book isn't a technical necessity. It's a choice.
Libraries, archives and museums already rely on mature non-destructive scanning technology. Google's own library partnerships used overhead cameras with infrared projectors to detect and computationally flatten page curvature rather than cutting bindings. The Internet Archive runs a similar non-destructive pipeline, and commercial robotic scanners like Treventus's ScanRobot can process up to 2,500 pages an hour while keeping the book intact.
The tradeoff is real, not imaginary. Destructive scanning produces perfectly flat pages with less gutter distortion, which can mean cleaner input for optical character recognition at scale. But that's an argument about speed and per-unit cost for pure text extraction, not a claim that preservation-friendly scanning doesn't work.
That distinction matters for how the FTC letter's core argument should be read. Bulk buyers choosing the destructive route aren't doing it because no alternative exists. They're doing it because destroying the book is faster and cheaper when the only thing that matters is the text, not the object.
The Anthropic Precedent Cuts Both Ways
The letter leans heavily on a 2024 lawsuit against Anthropic, where author Andrea Bartz and others accused the company of acquiring, scanning and discarding millions of print books.
That case already reached a verdict, and it's more complicated than the letter's framing suggests. A judge ruled in 2025 that Anthropic's use of legally purchased books to train Claude did not violate copyright law.
That distinction is worth sitting with. Anthropic won on the specific question of whether scanning legally acquired books to train an AI model infringes copyright.
The FTC letter isn't reviving that copyright question. It's arguing that permanently destroying scarce physical books harms competition and public access on its own terms, regardless of whether the scanning itself was legal.
Anthropic didn't immediately respond to CBS News's request for comment on the new letter.
What This Means for Miami
This adds a new front to the AI copyright fights MAIN has already tracked this year, from the Tung Tung Sahur meme character case to ByteDance's dispute with Hollywood's Motion Picture Association.
The distinction matters for any Miami business handling physical archives, rare documents or specialized print collections being courted by AI companies for training data. A legal win on copyright, which is what Anthropic already has, doesn't necessarily protect a company from a separate antitrust or unfair-competition theory if the underlying conduct destroys scarce, non-renewable material in the process.