Recent investigations into Amazon AI training practices reveal that the technology corporation has been purchasing and dismantling physical books to feed large language models. The discovery emerged after an investigative outlet placed a tracking beacon inside a rare volume sold in a bulk order to an anonymous buyer.
Tracking Operations at the Nevada Facility
An investigation conducted with a bookseller tracked a hidden AirTag to an Amazon facility in Las Vegas known as VGT3. The facility housed a specialized team tasked with separating book spines and scanning individual pages for data ingestion, according to Ars Technica. Reports indicated that the site bore a team logo depicting a dinosaur holding a book.
Amazon issued a formal response regarding the acquisition of these texts without directly detailing the scanning process. The company said that it purchases books through standard commercial channels to improve customer services and products.
The Search for Clean Datasets
Technology companies developing frontier systems face a shrinking supply of clean human-written text across the internet. Training modern systems requires original, non-synthetic data to avoid performance degradation known as model collapse. Rare physical works printed before 2022 guarantee that the text contains no computer-generated phrasing.
Consequently, development teams have turned to obscure physical prints, out-of-print titles, and foreign-language texts to expand artificial intelligence capabilities. Internal forum discussions reviewed by investigators showed that workers at the Las Vegas warehouse previously faced text inventory shortages that threatened operational continuity.
The Expanding Scope of Amazon AI Training
Workers assigned to the Amazon AI training team were instructed to systematically catalog international standard book numbers before processing each volume. This procedural step suggests an organized attempt to scan broad catalogs of distinct published ISBN records. Meanwhile, competing developers such as Anthropic and xAI have stated that they do not use antique or rare volumes for their projects.
The data collection process impacts the broader economy of independent booksellers who unknowingly supply these projects through third-party brokers. Industry observers note that high-volume orders have altered inventory patterns across the retail market.
Ethical Dilemmas for Booksellers and Historians
Archivists and antiquarian sellers argue that destructive scanning risks erasing unique historical artifacts that exist in limited surviving copies. Many older books contain marginalia, historical context, or niche academic records that have never been digitized for public preservation repositories.
In addition, privacy and asset tracking tools developed for consumer smart devices have now become instrumental in uncovering hidden supply chains within the technology sector. As tech leaders pursue advanced models, debates around cultural preservation and data sourcing transparency will continue to intensify.
