Scale.
The Latest
Tesla's Cybercab Hits a Federal Wall: The Regulatory Math Every Founder Building Autonomous Tech Needs to SeeOne App Said No to 'Lake America' - and Rocketed to No. 1Ford's $28,350 Fathom Is the Affordable Electric Truck Founders Actually Want to DriveMeta's $17 Billion Teen-Safety Settlement Has a Catch: TikTok and YouTube Have to Pay TooChipotle Pulls Jalapeños After 110 Salmonella Cases — Here's What Its Traceability Playbook Actually DidFIFA's $20 Billion Power Grab Collapsed in 72 Hours — Here's What Founders Can Learn About Opaque Deals4 Lounges, 1 Credit Card: Southwest's Playbook to Turn Loyalty Into RevenueDyson's $500 Toothbrush: The Pricing Playbook Behind Its Boldest Bet YetHow Olivia Rodrigo Turned One Festival Into a $20 Million Fundraising MachineAn AirTag Inside a Rare Book Led Straight to Amazon's AI Training WarehouseTesla's Cybercab Hits a Federal Wall: The Regulatory Math Every Founder Building Autonomous Tech Needs to SeeOne App Said No to 'Lake America' - and Rocketed to No. 1Ford's $28,350 Fathom Is the Affordable Electric Truck Founders Actually Want to DriveMeta's $17 Billion Teen-Safety Settlement Has a Catch: TikTok and YouTube Have to Pay TooChipotle Pulls Jalapeños After 110 Salmonella Cases — Here's What Its Traceability Playbook Actually DidFIFA's $20 Billion Power Grab Collapsed in 72 Hours — Here's What Founders Can Learn About Opaque Deals4 Lounges, 1 Credit Card: Southwest's Playbook to Turn Loyalty Into RevenueDyson's $500 Toothbrush: The Pricing Playbook Behind Its Boldest Bet YetHow Olivia Rodrigo Turned One Festival Into a $20 Million Fundraising MachineAn AirTag Inside a Rare Book Led Straight to Amazon's AI Training Warehouse
Scale.
Sections
The outlet
Technology

An AirTag Inside a Rare Book Led Straight to Amazon's AI Training Warehouse

A bookseller planted a tracker in a bulk shipment and watched it travel to a Las Vegas facility where workers cut spines and scan pages to feed Amazon's frontier models. Here's what founders building on AI need to understand about where the training data comes from.
AUG 19, 2026
Photo: Foto: arstechnica.com

The supply chain for AI training data just got a lot more visible — and a lot more controversial.

A rare bookseller suspicious that AI companies were bulk-buying books and destroying them planted an AirTag inside a shipment. According to a 404 Media investigation, that tag traveled to a warehouse in Las Vegas called VGT3, operated by Amazon. Workers there cut the spines off books to scan pages faster, then discard the originals. The team's logo, documented by 404 Media, shows a Tyrannosaurus rex devouring a book.

Amazon declined to confirm the books are used for AI training. Its statement said only that the company 'purchases books through commercial channels to help develop and improve the products and services our customers use.' That's it. No mention of AI.

But the context is hard to ignore. Amazon is building what it considers frontier AI models — the kind that compete directly with Google, OpenAI, and Anthropic. Those models need massive volumes of unique text. Rivals Anthropic and xAI have publicly stated they are not training on rare or antique books, which would give Amazon a potential data edge if it is.

The ISBN theory gets legs

404 Media also surfaced online forum discussions among VGT3 workers indicating they are trained to scan barcodes and ISBNs before scanning book pages. That practice, 404 Media reported, supports a theory booksellers have held for a while: AI companies are methodically working through ISBN lists to ensure every unique printed book gets scanned at least once.

Workers in those same forums reportedly said the facility nearly shut down earlier this year after running out of books to scan. The supply completely ran out at one point, they said. The facility is still operational.

What booksellers are actually losing

The books being targeted are not typically high-value first editions. They tend to be older works with lower monetary value — books never widely distributed, never translated into widely spoken languages, never popular enough to survive in large quantities. But the bookseller who planted the AirTag told 404 Media these works still carry 'historical value, intellectual value, sentimental value' that AI companies overlook. 'They just want the content as a bunch of words strung together,' the bookseller said, according to 404 Media.

One named critic, Burry, reportedly called the practice 'evil incarnate,' according to Ars Technica. Amazon workers, for their part, described the gig in forums as a 'nice' opportunity with flexible hours.

What this means for founders

If you are building products on top of any major AI model, the training data question is no longer abstract. The competitive moat these companies are digging is built partly on content that cannot be recreated — and that is being consumed in a one-way process. The knowledge gets locked inside proprietary weights, unavailable to competitors or researchers.

Free enterprise runs on property rights and voluntary exchange. Bulk-buying rare books through commercial channels is legal. But destroying irreplaceable works to feed a closed, proprietary system — while refusing to disclose what is being scanned — is the kind of opacity that should make any founder ask harder questions about the infrastructure they are building on.

EP
Ethan Park
Technology Columnist

Ethan Park covers AI, software and the tools reshaping how business gets done — always through the lens of what an owner should do about it.

More from Technology

Dyson's $500 Toothbrush: The Pricing Playbook Behind Its Boldest Bet Yet
Meta's $17 Billion Teen-Safety Settlement Has a Catch: TikTok and YouTube Have to Pay Too
This Services Firm Turned Itself Into a Product. Here Are the 4 Decisions That Did It.