Secondhand booksellers across the UK and Ireland are reporting unusual bulk orders from mystery buyers, with some suspecting that artificial intelligence companies could be acquiring physical books to use as training data.
The orders, which have included hundreds of unrelated titles, have generated thousands of pounds in sales, but the identities of the buyers and their reasons for purchasing the books remain unconfirmed.
The reports follow revelations about AI developers acquiring physical books to extract their contents. Anthropic, the company behind the Claude chatbot, was reported to have spent tens of millions of dollars on books for a scanning operation.
However, there is no confirmed connection between that project and the orders received by independent booksellers.
Unusual Bulk Orders Unexplained
Stuart Manley, co-owner of Barter Books in Alnwick, Northumberland, said the unusual orders began arriving about three months before the Guardian's August report.
Unlike conventional bulk purchases, which are usually organised around subjects such as sport or motoring, the requests contained seemingly unrelated titles.
'It's definitely a strange combination of books,' Manley said. 'Normally a book order is on a theme but this is all over the place.'
One order included an Estonian translation of John le Carré's The Mission Song, a particular edition of Anne Brontë's Agnes Grey and an October 1983 issue of the magazine Warship. Manley said he had sold hundreds of books to three buyers for approximately £4,000.
The pattern was not confined to one shop. Jim Shaughnessy, owner of MW Books in Claregalway, Ireland, said his business had received large and varied orders since May.
He described the selections as 'sporadic, presumably machine-driven' and said they had little thematic connection, ranging from books about agricultural equipment in 18th-century Africa to biographies of 1950s racing car drivers.
David Gower-Spence, owner of BookLovers of Bath, also reported a rush of orders between May and July, involving approximately 200 books. He said the volume had been overwhelming despite his business being relatively well organised.
An unnamed UK bookseller told the Guardian that some orders placed under different aliases appeared to be destined for the same address. The seller said they had received orders for approximately 6,000 books from similar buyers since January, generating thousands of pounds in sales.
The buyers were reportedly willing to pay full prices rather than negotiate discounts. Even so, the bookseller said the activity was disrupting the normal rhythm of secondhand trading.
The shared delivery address reportedly pointed to freight warehouses near Heathrow Airport, although the identities of the buyers and any relationship between them have not been established.
Anthropic's Book-Scanning Project Fuels AI Training Data Suspicions
The suspicion surrounding the orders comes after reports about Anthropic's Project Panama.
The Washington Post first reported in January that the company had spent tens of millions of dollars acquiring millions of physical books for an operation involving removing their spines, scanning their pages and sending the books for recycling.
The exact number of books and total project cost were redacted in court documents.
Anthropic said Claude was trained using publicly available web data, commercially acquired datasets and data generated by the company itself. It also described book sourcing as a widely used approach across the AI industry.
'None of our data acquisition programs buy and destroy rare or antiquarian books,' the company said.
That statement did not deny the previously reported acquisition and scanning of books. It specifically addressed rare and antiquarian titles, leaving a distinction between those works and ordinary secondhand publications.
The company's book acquisition has also been the subject of copyright disputes.
A US court's fair-use ruling on training with lawfully acquired books was separate from claims involving the acquisition and storage of pirated books.
Anthropic later agreed to a $1.5 billion (£1.11 billion) settlement with authors and publishers without admitting wrongdoing.
For booksellers, the possibility that older physical publications could be valuable to AI developers has added another explanation for the strange requests. Printed books may contain material that is difficult to find online, including older editions and works that have not been digitised.
But that remains a possible explanation, not proof of what is happening to the books being ordered.
Mystery Buyers Have Yet to Be Linked to AI Companies
Another story added to the speculation in July, when 404 Media reported that ISBNdb had advertised a service offering printed books as potential AI training material. Its marketing suggested that older publications could be attractive because they predated the widespread use of generative chatbots.
ISBNdb later removed the webpages and described the proposal as a test of market interest. The company said it had never purchased, scanned or sold books for AI training.
Some booksellers have also questioned whether AI training is a convincing explanation for every order. Requests have reportedly included public-domain works such as Jonathan Swift's Gulliver's Travels, first published in 1726 and available online for free.
One buyer mentioned by the booksellers was Canada-based Zoom Books, which describes itself as a book-recycling company. The Guardian contacted the business for comment, but the supplied report did not establish that it was acting for an AI developer.
The sellers' accounts point to a genuine change in the orders reaching their shops. What they do not reveal is who is ultimately acquiring the books, whether the different buyers are connected or whether any of the purchases are being used to train AI systems.
