Amazon says it buys books in bulk to ‘improve products’
Amazon has said it buys books through commercial channels to develop its products after an investigation traced a bulk order of rare books to a facility where workers reportedly destroy and scan physical copies.
“Amazon purchases books through commercial channels to help develop and improve the products and services our customers use,” an Amazon spokesperson told City AM.
Last week, an investigation by 404 Media tracked an order of around 1,000 books from an anonymous buyer to an Amazon facility in Las Vegas using an Apple Airtag hidden inside one of the volumes.
Workers at the site, known as VGT3, reportedly remove the spines of books before scanning their pages, leaving the physical copies unusable.
Employee posts cited by 404 Media also indicated workers scan the books’ barcodes or ISBN numbers as part of the process.
Amazon did not specify in its response to City AM which products the material was being used to develop, or explicitly confirm that the scanned books were being used to train its AI models.
It comes amid a growing scramble among tech companies for high-quality written material as they develop increasingly powerful AI systems.
Printed and out-of-print books can be particularly useful because much of their content is unavailable online. Older texts also pre-date the explosion of AI-generated material on the internet, which has complicated efforts by developers to source reliable human-written training data.
Amazon develops its own family of Nova AI models and is competing with the likes of OpenAI, Google and Anthropic in the increasingly expensive race to develop new AI models.
Following Anthropic’s lead
Anthropic, the developer of Claude, had already been found buying millions of second-hand books under an internal programme known as Project Panama before cutting off their spines, scanning the pages and recycling the originals.
A US federal judge subsequently ruled that Anthropic’s digitisation of legally purchased books amounted to fair use, distinguishing it from the company’s separate use of pirated copies.
Anthropic later agreed a $1.5bn (£1.13bn) settlement covering claims relating to pirated books.
City AM reported last month that rare book dealers in London had received unusually large approaches from anonymous buyers seeking specialist titles.
One bookseller said they received two sizeable enquiries from email addresses carrying no names or company details.
The prospective buyers stopped responding after being asked to identify themselves and explain where the books would go.
Similar reports from booksellers have raised concerns that AI developers or their suppliers are systematically seeking obscure and out-of-print titles which have not previously been digitised.
The race for fresh data is becoming a growing constraint for AI developers as the easily accessible internet has already been extensively scraped.
Publishers have meanwhile begun turning that demand into a new revenue stream: Amazon struck an AI licensing agreement with the New York Times last year, reportedly worth between $20m and $25m annually.