Someone Is Buying Japan's Used Books by the Ton, Shipping Them to America, and Not Explaining Why
Japanese used-bookstores are seeing five-times-normal sales as mystery buyers snap up philosophy, law, and history titles in bulk. Export records show at least 50 tons — roughly 100,000 volumes — have gone to a US facility. Nobody involved will say what happens next, though 'destructive scanning' is the industry's word for it.
Sometime around August, used-bookstores across Japan started noticing something odd: ordinary days turning into extraordinary ones, with sales running roughly five times normal volume, driven not by the usual manga and paperback novels but by a strangely specific slate — philosophy, history, political history, medicine, law, and documentation from Japan’s Edo period. The orders came through multiple buyer accounts, spread across different stores, all routed to the same place: a logistics center in Okayama Prefecture, whose operator has declined to comment on what happens to the books once they arrive.
Trade data tells the rest of the story that the buyers won’t. A group company tied to a major Japanese publishing distributor has exported more than 50 tons of material labeled simply “Japanese books” to the United States since 2025. At roughly 500 grams a volume, that’s an estimated 100,000 books, gone. As one industry observer put it, the pattern looks less like a reseller stocking up and more like “someone assembling a corpus” — the term of art for a dataset, not a library.
Nobody has confirmed what the books are for, and that matters, because it means everything that follows is inference rather than fact: no buyer has come forward, no scanning facility in Japan or the US has been publicly identified, and no company has been named as the recipient. What can be said is that the leading theory fits a pattern the AI industry already has a name for — destructive scanning, the practice of removing a book’s binding before running it through a high-speed scanner, which produces a clean digital copy and, not incidentally, a book that can no longer be resold, returned to circulation, or read by anyone who isn’t an AI model in training. The binding comes off, the pages get fed through, and what’s left goes in the shredder. It is, book for book, the most efficient way anyone has found to turn a library into a dataset, and the least efficient way anyone has found to keep a library a library.
If the phrase “destructive scanning” sounds oddly specific for readers of this publication, it’s because it isn’t new — it’s the same practice at the center of a real, ongoing US lawsuit against Anthropic, where the company was accused of buying and scanning millions of print books to build training data, tearing off covers and bindings in the process. That case involves Anthropic specifically; nothing has tied this Japanese shipment to Anthropic, or to any other named AI company, and it would be its own kind of fabrication to imply otherwise. But the method has a paper trail now, and the paper trail runs directly through what’s showing up in export data out of Okayama.
Japan’s copyright law, unlike America’s, has a specific provision built for exactly this scenario, and it’s more permissive than you might expect. Article 30-4 of the Copyright Act allows certain “information analysis” uses of copyrighted work without the rightsholder’s permission — training an AI model arguably qualifies — provided the use doesn’t “unreasonably prejudice the interests of the copyright holder.” Nobody has tested in court whether buying up a country’s supply of Edo-period medical texts, shredding them, and feeding them to a foreign AI company crosses that line, because nobody has been identified as having done it yet. The law was written with fair use for researchers in mind, not a bulk-logistics operation with unmarked buyer accounts and an unresponsive Okayama warehouse.
For the bookstores themselves, this is, in the most literal sense, a windfall — a category of merchandise nobody wanted, suddenly worth five times as much, bought by people who don’t haggle and don’t ask questions. One Tokyo bookseller told local reporters that novels and comics used to be the business; now it’s whoever’s paying premium rates for a stack of 1920s legal commentary nobody had touched in a decade. It is hard to begrudge a struggling used-book trade a genuine sales surge, even one this strange. It is also hard not to notice that the last time anyone saw this many philosophy and law books moving this fast, it wasn’t because interest in philosophy and law was suddenly booming — it was because someone, somewhere, needed philosophy and law books that nobody would miss.
What happens to a book once it’s shredded for scanning is not, legally or philosophically, the same question as what happens to the words inside it — the text survives, digitized, probably forever, inside whatever model eventually trains on it. What doesn’t survive is the object: the specific, physical, only-one-like-it copy of a 1920s legal text or an Edo-period medical scroll that a Tokyo bookstore had been quietly holding onto for however many decades, waiting for someone who wanted it enough to read it rather than feed it through a machine. Nobody involved has explained why that trade is worth making, mostly because nobody involved has agreed to explain anything at all.
Sources: