{"id":17192,"date":"2026-08-17T19:25:43","date_gmt":"2026-08-17T19:25:43","guid":{"rendered":"https:\/\/nile1.com\/en\/?p=17192"},"modified":"2026-08-17T19:25:54","modified_gmt":"2026-08-17T19:25:54","slug":"amazon-slices-and-scans-rare-print-books-in-las-vegas-facility-to-feed-ai-models","status":"publish","type":"post","link":"https:\/\/nile1.com\/en\/2026\/08\/17\/amazon-slices-and-scans-rare-print-books-in-las-vegas-facility-to-feed-ai-models\/","title":{"rendered":"Amazon Slices and Scans Rare Print Books in Las Vegas Facility to Feed AI Models"},"content":{"rendered":"

Machine learning data demands have driven Amazon<\/a> to buy thousands of out-of-print books specifically to scan into model datasets. The retailer adopted Anthropic<\/a>‘s methodology of physical digitization, executing a process that ends in the destruction of volumes holding documented historical market value.<\/p>\n

That destruction pipeline was exposed when investigative outlet 404 Media planted an active AirTag tracking device inside a lot of rare titles sold through specialized vendor Biblio. Location pings recorded the parcel traveling across air freight networks and trucking routes for weeks before ending at an Amazon fulfillment site in Las Vegas.<\/p>\n

Inside that Las Vegas facility operates a dedicated internal processing unit designated as VGT3. Warehouse personnel receive incoming physical shipments, feed bound volumes into industrial cutters to sever their spines, and process loose pages through high-speed optical scanners.<\/p>\n

Scanned text streams directly into training pipelines for Amazon’s Nova neural models, the architecture powering the company’s Alexa<\/a> assistant. With public web crawlers exhausting accessible digital text, engineers targeted pre-2022 print editions to avoid post-launch algorithmic content.<\/p>\n

AI development consumes roughly 10 trillion words annually, exhausting human-authored digital text repositories at a rate forcing reliance on physical archives. Projections indicate high-quality linguistic data reserves face complete depletion before 2026, driving tech firms to acquire offline print inventories rather than synthetic web copy.<\/p>\n

Antiquarian book dealers recorded sudden spikes in bulk inventory sales over the past twelve months. Corporate buyers purchased entire catalog lots without negotiating sales prices or asking about physical wear.<\/p>\n

Ingestion software automatically filters incoming inventory by rejecting any volume lacking an International Standard Book Number. Amazon’s procurement code systematically moves through global ISBN database registries, leaving unindexed pre-modern manuscripts and self-published books out of the acquisition pipeline.<\/p>\n

Restricting purchases to registered pre-2022 titles prevents AI-generated text from contaminating language training runs. Retraining neural networks on synthetic ChatGPT<\/a> text triggers structural model collapse, a mathematical degradation state where model output quality permanently collapses.<\/p>\n

Industrial spine-slicing duplicates acquisition tactics established by AI firm Anthropic to build training sets for its Claude model. Evidence of Anthropic purchasing physical volumes, cutting bindings, and digitizing text emerged during discovery in class-action copyright lawsuits filed by authors.<\/p>\n

Federal judicial rulings established that buying physical books, scanning their pages, and destroying the original paper can constitute fair use under U.S. copyright law. Physical destruction prevents duplicate commercial copies from entering circulation, neutralizing direct statutory reproduction claims.<\/p>\n

Amazon acknowledged operating the procurement pipeline across commercial retail channels to enhance consumer products and service tools. Spokespersons refused to disclose total book destruction counts or explain why mechanical destruction remains necessary after page capture.<\/p>\n

\n

Read also:<\/h3>\n