Amazon’s Las Vegas warehouse was reported to scan rare books and destroy them for AI training

Amazon’s Las Vegas warehouse operation was reported to be receiving large shipments of printed books, stripping off their bindings or spines, scanning the pages and destroying the physical copies in the process. The site was identified as VGT3, part of an Amazon warehouse complex in Las Vegas, Nevada, and the purpose of the work was described as feeding data into AI training systems. According to 404 Media, the investigation began with a rare book that reporters equipped with an Apple AirTag. The tracker led the shipment to Amazon’s Las Vegas facility. GeekWire later said the parcel moved from California to Milwaukee, spent time in a warehouse outside Kenosha, Wisconsin, then travelled by truck with an overnight stop in Grand Junction, Colorado, before arriving at an Amazon warehouse known as LAS8. Forbes described a similar route that ended in Las Vegas and said the Amazon books operation was coded VGT3. Workers quoted in the reporting described a split process. One said, “I work at VGT3 here in Vegas, and all we do is scan books.” The same account said some employees are assigned to cut books, while others receive them and scan the barcodes. NJ.com and TechCrunch reported that the VGT3 logo featured a dinosaur holding a book. Amazon responded by saying it purchases books through commercial channels to help develop and improve the products and services customers use. As covered last week, ISBNdb withdrew a similar AI book-scanning pitch after backlash. The broader pattern has already surfaced in litigation involving Anthropic. Forbes reported that court documents from that case showed the company bought millions of physical books, hired contractors to cut off bindings, fed the pages into scanners and shredded the originals in a project called Project Panama. Other reporting said Anthropic planned to scan between 500,000 and 2 million books over a single six-month vendor contract. That litigation also shaped the legal debate. Reports on the case said a judge found the conversion of lawfully purchased print books into digital files for internal model training to be transformative fair use because the digital copy replaced the original physical one. Coverage by Lit Hub and other outlets has since suggested that similar destructive scanning practices may have been used or contemplated by other large AI companies, including Meta, Google, Microsoft and OpenAI. The attraction of old books is straightforward in the reporting: they can provide large amounts of human-written text that is less likely to be contaminated by AI-generated material. TechCrunch said rare books, especially those out of print or unavailable online, offer a fresh source of training data, while other outlets noted that companies also value books printed before the current AI boom because they predate machine-generated text.