Anthropic says book sourcing is widely used; rare books not destroyed
The wave of bulk orders has reached secondhand booksellers across multiple continents. Bookshops contacted by the Guardian reported orders from buyers based in the US, Canada, continental Europe and the UK, with similar bursts documented in Australia and mainland Europe.
An anonymous UK secondhand bookseller told the paper that orders for 6,000 books from similar buyers since January had been worth thousands of pounds. The buyers, the seller said, are paying “top whack” and are not seeking a discount despite requesting titles in bulk. “It appears to be either a human or a machine dipping into the sandbox of secondhand books and what they are buying makes no sense in terms of the wider trends in the market,” the seller said. “It’s disrupting the secondhand bookselling business because this isn’t typical for the trade.”
The seller added that the aliases behind the orders are “very opaque even to experienced booksellers,” and that some orders appearing to come from different buyers are delivered to the same address. One delivery postcode seen by the Guardian and used by different buyers points to a group of freight warehouses near Heathrow airport.
Stuart Manley, who co-owns Barter Books in Alnwick, Northumberland, with his wife Mary, said the orders began arriving about three months ago. Manley described the selections as a strange combination of books, telling the Guardian: “It’s definitely a strange combination of books. Normally a book order is on a theme but this is all over the place.” A single order he received included an Estonian-language translation of John le Carré’s “The Mission Song,” a copy of Anne Brontë’s “Agnes Grey” from a specific imprint, and the October 1983 edition of “Warship,” a monthly magazine. Manley said he has sold hundreds of random books to three buyers for a total of about £4,000.
“My theory is that it’s masses and masses of AI money being squandered,” Manley said.
In Ireland, Jim Shaughnessy, owner of MW Books in Claregalway, said his business has been receiving large, varied orders since May. “They are sporadic, presumably machine-driven and have little in terms of thematic currency; books on agricultural implements in 18th century Africa through to biographies of 1950s racing car drivers,” he said. “Personally, I don’t see we have the right to dictate what any customer does with their purchases but it is more than curious to see what’s being selected.”
David Gower-Spence, the owner of BookLovers of Bath, said he received a rush of orders between May and July for about 200 books from buyers who have also been cited by other sellers contacted by the Guardian. “It was overwhelming for a while and I’m relatively well organised,” Gower-Spence said.
One of the buyers cited by multiple sellers is the Canada-based Zoom Books, which describes itself on its website as “North America’s leading book recycling company.” Zoom Books has been contacted for comment by the Guardian.
Similar bursts have been reported by booksellers in Australia and mainland Europe. The Guardian reported this month that Australian booksellers had also received a burst of orders, with one seller telling the paper the sudden scale of demand was “driving me nuts.”
Orders are typically placed through book marketplaces including Biblio and Amazon-owned AbeBooks. Both have been contacted for comment by the Guardian.
Asked about the bulk orders, Anthropic, the AI company behind the Claude chatbot, said: “Claude is trained on a mix of publicly available web data, commercially acquired datasets, and data we generate ourselves. Sourcing books is a widely used approach for training large language models across the AI industry. None of our data acquisition programs buy and destroy rare or antiquarian books.”
According to the Guardian, the Washington Post reported earlier this year that Anthropic had spent tens of millions of dollars acquiring books and slicing off their spines so their contents could be scanned before the books were sent for recycling.
The Guardian also reported last month that the tech site 404 Media had reported that a books database, ISBNdb, was telling booksellers that “the world’s best AI training data is sitting on a shelf,” and that books published before 2022 were ideal material for AI companies because their text was unpolluted by material that could have been generated by chatbots. ISBNdb said the report was based on a webpage that was a “test of market interest” and had since been taken down.
Booksellers have continued to debate the orders in online forums. One post on a bookseller forum, seen by the Guardian, said: “Shame these are all destined for AI training, and will therefore get the covers stripped and ultimately end up being pulped.” Other posts expressed puzzlement at the erratic selections, while some questioned whether the orders could really be for AI training when they included titles such as Jonathan Swift’s “Gulliver’s Travels,” first published in 1726 and long available free online.