Old and rare book sellers noticed an unusual uptick in large orders over the past year with common characteristics: apparently unrelated titles, enormous volumes and buyers showing little price sensitivity. According to 404 Media, suspicions pointed to artificial intelligence companies seeking texts to train their models.
To identify the buyer, the publication provided a bookseller with an Apple AirTag, which was placed inside a copy of an order of approximately 1,000 books made through the Biblio platform. The tracker revealed a route spanning thousands of kilometres: the shipment passed through California, Wisconsin, Colorado and finally arrived at LAS8, an Amazon warehouse in Las Vegas.
At the northern end of LAS8 operates a unit identified internally as VGT3, according to testimony from employees consulted by 404 Media. Workers describe a specific process: they receive the copies, register their barcodes or ISBNs and prepare them for digitalisation. The procedure includes removing or cutting the binding to scan pages more quickly, destroying the physical copy in the process.
Amazon did not explain what specific systems receive the digitised texts after scanning. The company stated only that it purchases books through commercial channels to develop and improve products and services. The texts acquired include material that is not readily available on the internet and, in many cases, was published before the current expansion of synthetic content.
The discovery coincides with previously documented operations. Anthropic executed a similar process during the so-called Panama Project, purchasing millions of printed books that were converted to digital copies through destructive digitalisation. A judge considered that procedure fair use when applied to legally purchased copies, although it reached different conclusions regarding books downloaded from illegal libraries.
