I order a lot of secondhand and vintage books online. This year the prices jumped and I chalked it up to inflation. I was wrong.
The reason behind the spike in prices is what AI companies euphemistically dub "destructive scanning." That is when they buy millions of second-hand books in bulk and send them to enormous warehouses across America. There, automated assembly lines cut the pages from the spines and then feed the pages into huge industrial scanners.
Those books are being harvested for their text to expand the database for the LLM models. The physical pages? Later discarded. Some of these are the last remaining editions of historical and collectible works — once they're cut and scanned, they're gone.
Booksellers across the UK, Germany, New Zealand, Australia and the U.S. are reporting the same thing: bulk, random, non-fiction orders. Orders have ranged from 1,000 books to as many as a million in a single purchase. Just one vendor contract called for scanning 500,000 to 2 million books.
Anthropic ran this under an internal name, Project Panama, and tried to keep it quiet. Some booksellers noticed something was odd. A Devon bookseller, for instance, got an odd order this summer for books on subjects as disparate as Cornish pioneers in South Africa, cheeses of the world, and Hong Kong jewelry — from a "shipping company" near Heathrow. Her suspicions grew fast. It turned out to an AI run office.
I should have seen this coming. I'm one of the author-plaintiffs in the Bartz et al. v. Anthropic class action — a different kind of theft. Anthropic had trained Claude on more than seven million digitized books, sourced from pirate websites, without paying a penny to authors or publishers. Mine included, seven of my thirteen. There are 120,000 individual authors and rightsholders (representing 447,576 eligible works out of 482,460 total listed books).
A federal judge approved the $1.5 billion settlement this summer. It was the court documents from that litigation that exposed Project Panama. I missed it until the story recently hit the Washington Post and Daily Mail and other outlets.
Having learned an expensive lesson from its class action payout, Anthropic is now going after physical books a different way — and legally, this time. Under the first-sale doctrine, established by the Supreme Court in 1908, once someone buys a physical book, they can do whatever they want with it, including destroying it.
So Anthropic attorneys have determined that buying and scanning a physical copy sidesteps the piracy problem. And once that book is shredded, any future Claude user can draw on its contents for free — the original gone forever.
Some call this a modern-day book burning. The Nazis destroyed books to erase what was in them — these companies destroy books to steal the information and then essentially sell it back to you. I am still angry this is somehow legal.