Slop

The impact of AI is no longer only in digital terms, but it affects frameworks as seemingly little linked to the virtual environment as second-hand books. Booksellers around the world are detecting the purchase of batches of books focused on very specific points. By crossing the data, the intentions are guessed, and again the AI ​​and its voracious need to be fed with data in an almost gargantuan way is behind it. Buyer robots. Marçal Font runs the Fènix bookstore, in Badalona, ​​and has been receiving purchases for weeks from a Canadian company that suspects it is automated: consecutive orders from the same buyer separated by just a minute. About twenty Spanish second-hand bookstores have sold to the same company since the end of April, some with orders of more than a thousand copies, almost always out-of-print non-fiction: they talk about a monograph on the castellers of Granollers from the seventies, a technical winemaking manual, conference minutes from half a century ago, diaries from the Civil War… All over the world. The same pattern is repeated in Germany, the United States, New Zealand and Australia: a Silicon Valley company has placed an order for more than 3,000 books with a Spanish bookstore. What distinguishes this type of buyers from ordinary customers is the indifference about price and the absence of thematic coherence. another example: An American bookseller went from selling twenty books in a good week to hundreds, with purchases that ignored the market value of the copy and an eye for subtlety, jumping from one topic to another with no apparent relationship. On the forums of Alibris, another book-selling marketplace, a platform official attributed the uptick to the arrival of new wholesale buyers who are hoarding commercial books. Two companies, one destination. The buyer identified in Spain is ZoomBooks, a Canadian second-hand book buying and selling company that sends orders to an Illinois logistics operator, PrepFort, in charge of scanning and cataloging the shipments. When asked, ZoomBooks denies collaborating directly with Anthropic, the company behind AI assistant Claude, and defines itself as a second-hand bookstore dedicated to recycling. Regarding the final destination of the books, and about the publications that its own website published about how to feed algorithms with used copies and destroy them later, the company responds that it does not comment on its commercial agreements because they are subject to confidentiality. Photo by Ugur Akdemir on Unsplash In parallel, another type of intermediary operates, more explicit about its function: ISBNdb, a bibliographic metadata database that now offers AI laboratories the purchase of between a thousand and a million books on request. Your own website puts it crudely: printed books before 2022 are the reserve of human text that no internet tracking can any longer guarantee, protected against contamination by texts generated by AI itself that now flood the network. Forced destruction. This commercial offer is supported by a specific ruling. Federal Judge William Alsup ruled in June 2025, in the case Bartz v. Anthropicthat training language models with legally purchased books constitutes a transformative use protected by the doctrine of fair use. He also considered it legal to digitize those copiesbut only because the printed copy disappeared in the process. That is, the digital copy takes the place of the paper book. Put another way: if the physical book survived the scan, the company that bought it would keep two copies having paid for one, and the argument of fair use would weaken. The shredder is a mandatory step in the legal process. Jason Leung on Unsplash Secret project. Months before that ruling, Anthropic was already carrying out a large-scale purchasing operation known internally as Project Panama. An internal company document, uncovered by The Washington Post based on declassified judicial documents, thus defined the action: “Scan Destructively all the books in the world.” The same document asked that the project remain hidden. ISBNdb has turned that same caution into a sales argument: it offers its clients, AI companies, a confidentiality agreement with each order and bluntly admits the image problem caused by the mass destruction of books. No headline about destroyed books generates sympathy, or as they themselves say, “destroying books gives a bad image“. What worries booksellers. Spanish second-hand booksellers have alerted the Ministry of Culture of what is happening. Miguel Ángel Ortega, president of the UNILIBER antiquarian booksellers association, recalls that his union fulfills functions of preservation and conservation of bibliographic heritage in addition to buying and selling, and it is contradictory for him to sell copies only for them to end up destroyed. Font puts it without half measures: “we are facing a form of literary plunder.” What is already protected by legal deposit and well cataloged runs less risk. But there are also “fanzines, neighborhood newsletters, local publications, documentation of social movements…” according to Font. Xavier Vinaixa, a researcher who helped uncover the case, and analyst Antonio Ortiz agree that the industry increasingly needs human text not contaminated by AI to avoid the collapse of the model, although Ortiz clarifies that the scarcity of data weighs less today than it did two years ago because training relies increasingly on reinforcement learning on already qualified data. Researcher Patrícia Ventura talks about how knowledge has to remain a public good, not a private reserve of a handful of companies. Image | Prateek Katyal in Unsplash