When American artificial intelligence companies began purchasing books in bulk, Europe was the most sensitive region, and it was also where the massive orders of used books were first reported. Were those ‘abnormal orders’ actually placed by an AI company? Two shipments of used books bound for a book-scanning companyβthe process of converting paper text into digital filesβwere discovered in Europe. This analysis was conducted on the 18th using ImportKey, a platform providing import and export shipping data.
The Shipping Trail from Europe to the United States
According to a Bill of Lading (a document issued by a carrier acknowledging receipt of cargo) found on ImportKey, a cargo shipment arrived in Norfolk, Virginia, on February 23. The shipment originated in Rotterdam, Netherlands, and was initially received in Grangemouth, United Kingdom. The interesting parts were the seller and the buyer. The seller was the UK branch of ‘Better World Books’ (BWB), a company contracted by the US AI firm Anthropic for book procurement and delivery, while the buyer was Datamation, the company entrusted by Anthropic with book scanning. The cargo description on the bill of lading read ‘Used Books, Books,’ with a weight totaling 13.748 tons.
A cargo shipment that arrived in the US on April 30 was similar. The cargo description stated ‘Used Books,’ and the seller and buyer were once again BWB and Datamation. Only the port of departure differed, originating from Antwerp, Belgium, weighing 12.703 tons. Combining the shipments from February and April, assuming an average weight of 400g per book, this amounts to roughly 66,000 books. In summary, both shipments involved Anthropic’s book procurement partner sending used books to the book-scanning company contracted by Anthropic.
AI Training Process Revealed Through Lawsuits
The AI training process involving books deployed by Anthropic, as revealed during the ‘Authors v. Anthropic’ lawsuit, followed the exact same method. Once BWB collects and sends the books to Datamation, Datamation slices the book spines using hydraulic guillotines, scans them, transfers the digital files to Anthropic, and disposes of the physical books. Documents submitted to a US court show that Datamation stated it could scan between 500,000 and 2 million books over a six-month period. Anthropic converts the files received from Datamation into tokensβthe minimum unit of AI processing dataβto train its AI models.
Since the beginning of the year, suspicions have grown in Europe that AI companies were hoarding used books as a series of ‘abnormal bulk used book orders’ continued to surface. In August, an interview with a used bookstore owner was published in the UK daily newspaper The Guardian. The owner revealed they received an order starting in January to purchase 6,000 books, noting that despite the bulk purchase, the buyer paid top dollar without requesting any discounts. The owner remarked, ‘This is not standard practice in the industry.’ Additionally, German antiquarian bookstores were suddenly hit with a surge of large orders during early morning hours, and a used bookstore in Ireland received an order for specialized texts such as investment guides explaining Ireland’s Special Savings Incentive Account (SSIA) from some 20 years ago.
Global Bulk Book Inquiries Linked to AI
Recently, inquiries suspected to be backed by AI companies for bulk used book purchases have also emerged in South Korea. In August, a US firm, company B, contacted used bookstores operated by domestic online bookstores Yes24 and Aladdin via email, stating it wanted to purchase used books by the container unit.
