Wellington bookshop's bulk orders spark suspicion of AI training data buying

Wellington bookshop's bulk orders spark suspicion of AI training data buying

Anne-Marie Thorby, owner of the secondhand bookstore Book Haven in Wellington's Newtown, started getting unusual orders at the end of April. A buyer in Canada was sending multiple orders for books, sometimes minutes apart, and the volume kept growing until one single order asked for 71 books at once, more than the shop could physically fill. The requests all targeted niche, non-fiction titles specific to New Zealand: local history, a tourist pamphlet on Waitomo Caves, Maori language books, New Zealand cookbooks, biographies, political cartoons, even books from a local anglers club. The buyer specifically wanted books carrying ISBNs from the 1970s and 1980s.

The economics did not add up. A $10 book would end up costing $40 once postage and tariffs were included, leaving no way for a reseller to turn a profit. Thorby says that is what made her suspicious enough to start digging, and she raised the pattern on AbeBooks' online forum. There she learned the same thing was happening to other bookshops around the world, not just hers: shops in India and Europe reported the same kind of bulk buying. She says tens of thousands of books were being bought by one company and shipped to a warehouse in the United States.

Thorby's explanation is that AI companies are buying up secondhand books to train large language models, and in some cases destroying the physical copies as part of digitizing them, which she calls the only way for them to do it. As an illustration of the practice, she points to a line from Anthropic's court case: "Project Panama is our effort to destructively scan all the books in the world," a quote she attributes directly to the company. The article does not say Anthropic was the buyer placing orders with Book Haven; Thorby cites the Anthropic quote as an example of what she believes AI companies in general are doing, not as an identification of who was ordering from her.

After cancelling the first wave of orders, Thorby got another flurry the following week: 10 orders in a single day totaling 180 books, this time going to a different company than the original buyer. She cancelled those too. She describes the pattern as an anathema to booksellers, saying their role is to keep books alive and in the world, not to feed private profit by transferring knowledge into LLMs. Her advice to other bookshop owners facing the same kind of orders: don't do it, don't feed the machine, pause and think.

Key facts

  • Book Haven, a secondhand bookstore in Wellington's Newtown, began getting bulk orders at the end of April from a Canadian buyer, for niche 1970s-1980s New Zealand titles such as Maori language books, local history and a Waitomo Caves tourist pamphlet.
  • One single order asked for 71 books; owner Anne-Marie Thorby says a $10 book would cost $40 once postage and tariffs were added, making resale unprofitable and pointing instead to AI training data buying.
  • An AbeBooks forum discussion showed other bookshops in India, Europe and elsewhere in New Zealand seeing the same pattern, with tens of thousands of books reportedly going to one company's US warehouse.
  • A second flurry the following week brought 10 orders in one day for 180 books, this time from a different company; Thorby cancelled both waves of orders.
  • Thorby cites a quote from Anthropic's court case, "Project Panama is our effort to destructively scan all the books in the world," as an example of the practice, without identifying Anthropic as the buyer who ordered from her shop.

Why it matters

The story puts a concrete, first-hand face on a practice that usually stays invisible: bulk purchasing of physical books, on a scale that made no commercial sense, that Thorby suspects is feeding AI training pipelines. It surfaces alongside public knowledge of Anthropic's own disclosed "Project Panama" effort to scan books for training data, giving a bookseller's account of what large-scale acquisition of training material looks like from the seller's side.

Who it affects

Secondhand booksellers, who are the ones fielding these orders and absorbing the disruption; book preservation advocates and authors, whose out-of-print, niche and regional titles are the ones being targeted; and AI companies, whose data-sourcing practices are put under an unflattering, if partial, spotlight.

How to use it

Thorby's practical signal for other bookshop owners: watch for sudden, rapid bulk orders for niche or regional out-of-print titles with older ISBNs, especially where postage and tariffs make the order unprofitable for a reseller. Her advice once that pattern shows up is direct: don't fill the orders, don't feed the machine, pause and think.

How solid is it

The account rests on one first-hand source, Thorby, speaking to RNZ's Midday Report, corroborated only informally by other booksellers describing similar patterns on an AbeBooks forum. The specific numbers she gives, the 71-book order and the later 10 orders for 180 books, come directly from her; the broader "tens of thousands of books" and the identity of the buying company are her own characterization, not independently confirmed.

Risks and caveats

The buyer or buyers are never named, so it is not established which company or companies were ordering from Book Haven. The Anthropic "Project Panama" quote is offered by Thorby as an illustrative example from a separate court case, not as evidence that Anthropic itself placed these orders. No total dollar figures, no response from AbeBooks, the buyer, or any AI company, and no description of how the books would actually be converted into training data are given in the source.

“Project Panama is our effort to destructively scan all the books in the world.”

— Anthropic, quoted from its court case by Book Haven owner Anne-Marie Thorby