In brief
Second‑hand booksellers worldwide are seeing a massive uptick in orders that appear to be aimed at training AI systems. The rise coincides with a 2025 U.S. court decision that declared such use legal under copyright law. Internal documents reveal a company‑led program called "Project Panama" that scans and destroys books, sparking ethical debate. Booksellers grapple with the dilemma of selling rare texts that could be written off while finding profit in surplus supply.
The second‑hand book market has taken off in recent months as independent sellers report rockets of bulk orders that appear to be aimed squarely at AI developers. The sudden wave of sales has surprised shop owners in the UK and beyond, who are urged to handle the uptick carefully and keep their books in good condition while negotiating with buyers who may not actually require them for resale.
Stuart Manley of Barter Books in Northumberland said he has never seen anything on this scale—one Canadian client’s order equaled a week’s worth of sales. “I’ve sold two or three thousand books a week for the past 30 years. This single order was the same volume, just in one go,” he told us.
The surge aligns with a landmark U.S. court ruling in 2025. Judge William Alsup declared that using books to train AI software was “exceedingly transformative,” a finding that overturned the view that publishers’ books could be freely used for that purpose. The decision followed a lawsuit by three authors against Anthropic, a prominent AI firm, who had been buying and apparently destroying books in the process of training its chatbot, Claude.
Court documents patiently revealed an internal operation dubbed "Project Panama"—a project aimed at "destructively scanning all the books in the world." The program removes a book’s spine to ink the pages instantly, leaving a pile of recyclables. Though Anthropic’s spokesperson insists their data acquisition is standard practice, the fact that they are literally pulping books draws a sharp ethical line.
Bookshops across the globe respond in kind: some welcome the additional income, especially for titles that stagnate on the shelf. Others worry about losing unique, one‑of‑a‑kind editions that carry invaluable historical weight. The debate is further complicated by the UK’s stricter intellectual‑property laws, which require publishers’ permission for any copies used in training.
What do these trends mean for the future of reading? For one thing, the big‑data appetite of AI could be reshaping how books circulate. It also raises a question for collectors and historians: are we willing to sacrifice a single parchment‑bound copy for the sake of a high‑speed algorithm?
If you want to stay updated on how technology intersects with culture, consider subscribing to Tech Decoded.
