Topline
Authors, archivists, historians and bibliophiles have launched an online war against tech giants and artificial intelligence developers after an increasing mountain of evidence shows they’re buying rare and antique books en masse and feeding them to large language models—only to slice off their spines and destroy the original print copies in the process.
Key Facts
Documents made public in a lawsuit from authors against Anthropic earlier this year revealed the company had bought millions of physical books, hired contractors to slice their bindings off, fed the pages into high-speed scanners and then shredded the originals in a venture called “Project Panama.”
Reporting then revealed a historic spike in the sales of physical books printed before 2022, and booksellers began voicing their suspicions that AI companies were secretly their big buyers.
Book lovers and AI skeptics called it “evil incarnate" and railed against the practice, but more reporting from 404 Media and others—including an Apple AirTag tracking investigation—has revealed it’s not just physical bestsellers hitting the chopping block, but rare, antique and out-of-print books that may not be otherwise preserved.
404, a media company founded by tech journalists, on Monday published the results of an investigation in which an unnamed rare books seller, who’d received an order for almost 1,000 physical copies, allowed reporters to place an Apple AirTag inside one of the books.
The company tracked the package from California to Milwaukee, Colorado Springs and ultimately an Amazon-owned warehouse, coded by the company as “VGT3.”
The entrance to the building boasts a logo of a Tyrannosaurus rex preparing to eat an open book, 404 Media said, and employees report "nice" jobs where "all we do is scan books" all day.
The investigation has revealed how major tech giants are operating dedicated, industrial-scale book-scanning facilities in order to train large language AI models (LLMs).
WHY ARE AI COMPANIES BUYING PHYSICAL BOOKS?
AI companies have turned to using physical books because digital sources to train LLMs caused two major problems: legal liability and data quality. Using pirated online book archives, or "shadow libraries,” has exposed AI firms to massive copyright lawsuits from authors, artists, photographers and news outlets. But buying physical books legally entitles the owner to use that specific copy however they see fit—including to destroy it. The practice also gives access to millions of out-of-print, niche or specialized books that exist only in physical form. By buying books printed before 2022, companies can also guarantee they’re using text free from AI interference to train their models. The internet is now full of pages written by AI, and training new models on that text won’t allow it to become any smarter. But physical books offer human-authored (and highly edited) language, long-form coherent reasoning, narrative structure and other qualities short-form web pages and AI-generated text do not.
IS IT LEGAL?
Yes. First-Sale Doctrine of the Copyright Act is what allows the buyer of a physical book to destroy it if they so please. Other lawsuits decided that converting a lawfully purchased print book into a digital file for internal model training is a "transformative" fair use. Critics have argued companies are using physical ownership as a legal loophole to manufacture digitized datasets without compensating creators, but the practice remains legal.
HOW MANY BOOKS HAVE BEEN DESTROYED?
Nobody knows. Court documents unsealed during the Anthropic case revealed the company planned to scan between 500,000 to 2 million books over a single six-month vendor contract. Internal documents said it planned to "destructively scan all the books in the world." 404 Media sources reported anonymous orders ranging from 1,000 to 1 million books per purchase. Used booksellers in the U.S., U.K., and Europe report bulk orders for seemingly random non-fiction and out-of-print books routinely range from hundreds to thousands of titles per transaction.
CRUCIAL QUOTE
“Project Panama is our effort to destructively scan all the books in the world,” an internal Anthropic planning document said. “We don’t want it to be known that we are working on this.”