AirTag Tracks Rare Book to Amazon Facility Linked to AI Training

A tracking device hidden inside a rare book reportedly led journalists to an Amazon facility in Las Vegas where workers scan and dismantle books for AI-related purposes.

The investigation raises fresh concerns about how technology companies source training material for artificial intelligence. Critics fear that rare or difficult-to-replace books may be destroyed during the digitisation process.

AirTag tracks book to Amazon facility

404 Media worked with a bookseller to place an AirTag inside a rare book included in a bulk order.

The tracker later showed that the shipment arrived at an Amazon facility in Las Vegas. According to the report, workers at the site prepare physical books for scanning by removing their bindings and separating the pages.

Amazon did not directly address the investigation’s specific findings. However, the company said it buys books through commercial channels to help develop and improve products and services for customers.

The report did not establish exactly which AI product may use the scanned material or how Amazon processes the resulting digital copies.

Amazon AI book scanning sparks preservation concerns

Scanning an intact book can take time and requires specialised equipment. Removing the spine allows workers to feed individual pages into high-speed scanners, making the process faster and cheaper.

That method may work for common copies, but it has prompted concern when rare books are involved. Once a book has been cut apart, it cannot return to circulation, be resold or be preserved as a physical historical object.

The investigation also highlighted online discussions attributed to employees at the Las Vegas warehouse, known as VGT3. Those posts suggested that workers scan books, process their pages and discard the physical copies afterwards.

The facility’s logo reportedly features a Tyrannosaurus rex destroying a book, adding to the criticism surrounding its work.

ISBN scanning could help locate unique texts

Employees allegedly discussed scanning barcodes and ISBNs before processing books.

ISBNs identify specific book editions and can help buyers avoid purchasing duplicate copies. In an AI training context, they could also help companies locate unique material that has not already been digitised or widely published online.

This has led critics to question whether AI companies are attempting to build massive collections of printed texts by systematically sourcing books through ISBN listings.

AI developers need large amounts of written material to train language models. Rare books, old editions and niche publications could provide content that is not available in modern digital libraries.

Destroying purchased books remains controversial

Amazon is not the only AI company to face scrutiny over the use of physical books for model training.

Reports last year said Anthropic acquired and destroyed millions of books while building training datasets for Claude. Companies can generally alter or dispose of copies they legally purchase, but copyright questions may still arise around creating and using digital reproductions.

Some observers argue that destroying a purchased copy is legally safer than retaining a scanned duplicate. Others say the focus should remain on preserving rare and culturally important books, especially when non-destructive scanning methods exist.

The Amazon AI book scanning report shows the growing tension between the demand for AI training data, copyright law and the preservation of physical books.