AI firms are quietly buying up secondhand books by the thousands. Some are being destroyed to train chatbots.
A US court ruling cleared Anthropic to use purchased books for AI training. Now booksellers report a strange surge in sales of obscure titles, and court documents reveal a project with a goal to 'destructively scan all the books in the world'.
Key points
- A US judge ruled in 2025 that Anthropic, the AI company behind the Claude chatbot, could legally use purchased books to train its AI without violating copyright law.
- Internal Anthropic documents, unsealed in 2025, revealed a programme called "Project Panama" with a stated aim to "destructively scan all the books in the world".
- Destructive scanning physically removes a book's spine so pages can be scanned rapidly, after which the remains are recycled.
- Secondhand booksellers across the UK report a sharp rise in bulk purchases of obscure titles, with buyers giving no explanation for their orders.
- UK copyright law, unlike US law, requires the permission of the copyright holder before books can be copied for AI training.
Something odd is happening in the secondhand book trade. Titles that sat unsold for two decades are suddenly flying out the door. Obscure Latin texts. Cowboy novels nobody asked for. Academic monographs printed in runs of a hundred copies.
Neil Manley, who runs Barter Books in Northumberland, told BBC Technology he can't explain the surge. The sales are random, the buyers anonymous. The pattern points in one direction: AI companies buying books in bulk to feed their systems. We first reported on booksellers raising this alarm on 12 August, in "AI Companies May Be Destroying Rare Books to Train Their Models".
What is actually happening to these books?
Some are being destroyed. Court documents unsealed in 2025 from a lawsuit against Anthropic, the company behind the Claude chatbot (a large language model, the technology that powers AI assistants like ChatGPT), revealed an internal project called "Project Panama." Its stated goal was to "destructively scan all the books in the world."
Destructive scanning means exactly what it sounds like. A book is shipped to a facility, its spine is cut off so the loose pages can run through a high-speed scanner, and what remains is recycled. The book is gone.
| Programme detail | What the documents say |
|---|---|
| Internal name | Project Panama |
| Method | Destructive scanning (spine removed, pages scanned) |
| Stated goal | "Destructively scan all the books in the world" |
| Books acquired via | Purchase on open secondhand market |
| Rare/antiquarian books | Anthropic says it does not buy these |
An Anthropic spokesperson said Claude is trained on a mix of publicly available web data, commercially acquired datasets, and data the company generates itself. "None of our data acquisition programs buy and destroy rare or antiquarian books," they added, calling bulk book purchasing a widely used approach across the AI industry.
That last point is almost certainly true. Anthropic isn't alone here.
Why would an AI company want obscure old books?
Because weird, rare material is valuable training data. Large language models, the software engines inside tools like Claude and ChatGPT, learn by processing enormous quantities of text. Common books and websites have already been scraped repeatedly. Unusual texts, titles no algorithm has seen before, can sharpen a model's grasp of language and history in ways that another copy of a bestseller simply cannot.
Professor Emily Hudson, an intellectual property specialist at Oxford University, notes that the legal picture differs sharply on either side of the Atlantic. Judge William Alsup ruled in 2025 that Anthropic's use of the authors' books was "exceedingly transformative" and therefore permitted under US law. Professor Hudson's position on the UK is blunter: all copying for AI training starts from a requirement of the copyright holder's permission, full stop.
Should booksellers be worried?
The mood among dealers is mixed, not panicked. David Tobin of Walden Books in north London welcomes the trade but says it "would be sad" if the books end up destroyed. Derek Walker, owner of McNaughtan's in Edinburgh, draws a line between expendable copies and irreplaceable ones. A rare 18th-century edition that's survived this long deserves better than a scanner blade.
Manley is pragmatic. "The world no longer needs five million copies of The Da Vinci Code," he says. For surplus titles, recycling via an AI pipeline isn't obviously worse than landfill. The real problem is that nobody buying these books is saying what they plan to do with them. That's the thing worth watching: not whether the trade continues, but whether anyone ever has to disclose it.
Common questions
Is it illegal for AI companies to buy and scan secondhand books?
A 2025 US court ruling found Anthropic's use of purchased books for AI training was legal. UK copyright law is stricter: it generally requires the copyright owner's permission before text can be copied for any purpose, including AI training.
How can I tell if a book I'm selling might end up being destroyed?
You almost certainly can't. Buyers running these programmes don't identify themselves or state their intentions. If preserving a specific title matters to you, donating it to a library or archive is the safest route.
Are all secondhand books at risk, or just certain types?
AI buyers are drawn to unusual texts that provide material a model hasn't encountered before. Common bestsellers are less likely targets, though nothing in the current system stops bulk buyers from purchasing any title on the open market.
