Anthropic Cut the Spines Off Millions of Books to Train Claude

A warehouse full of books is now one of AI's most valuable resources.

Anthropic bought millions of physical books, sheared off their spines, scanned every page, and sent the pages to recycling.

Court filings unsealed in January revealed the operation, codenamed "Project Panama."

A federal judge had already ruled the destructive scanning legal under the U.S. Fair Use Law.

New reporting from 404 Media shows the practice spreading, with brokers helping AI firms buy at an industrial scale while keeping their clients unnamed.

"The optics problem is real. 'AI company destroys two million books' is not a headline that generates sympathy," ISBNdb wrote in its blog while describing its bulk sourcing service for AI developers.

The courts have settled the copyright question, but authors, booksellers, and collectors are now arguing about a different one.

Their concern is what happens when the last copy of an out-of-print title goes through a hydraulic cutter.

Piracy Charges Sent Buyers Back to Print

Anthropic Co-Founder Ben Mann downloaded millions of pirated books from Library Genesis and Books3 in 2021, according to court records.

The legal exposure from this library pushed the company toward paper.

In February 2024, Anthropic hired Tom Turvey, the former head of partnerships at Google Books.

An internal planning document put the goal in writing: "Project Panama is our effort to destructively scan all the books in the world."

Hydraulic cutters removed the spines, and high-speed scanners captured every page before the paper went to recycling.

Anthropic spent tens of millions of dollars on the books, logistics, and scanning.

Judge William Alsup ruled in Bartz v. Anthropic that scanning legally purchased books counted as transformative fair use.

Why? Because each physical copy was replaced with a single digital copy. 

The first-sale doctrine did the rest, since a buyer can do what they want with a book they own.

But Judge Alsup drew a hard line at the more than seven million books Anthropic downloaded from pirate libraries.

He found this acquisition illegal on its own terms, and settling it cost Anthropic $1.5 billion.

Alsup retired before the settlement closed, and Judge Araceli Martinez-Olguin granted final approval, covering roughly 482,000 works at about $3,000 each.

Elon Musk moved to separate his own AI operation from the practice.

"I've asked the SpaceXAI team to preserve any rare books in a library and scan them the hard way vs just cutting off the spine and scanning," he wrote on X.

Tech founder David Sacks accused Anthropic of hypocrisy for training free on other people's work while treating its own output as protected IP.

Meta, OpenAI, and Google are defending their own training-data suits, and the Bartz ruling handed all of them a lawful supply route.

The catch is that the cheapest path to clean training data also produces the worst possible headline for the brand behind it.

Rare Books Enter the AI Supply Chain

Online book database ISBNdb spent years helping bookstores and libraries find titles.

The company now brokers orders of 1,000 to one million books for AI developers, with an NDA on every engagement.

The pitch came down to the argument that books printed before 2022 carry no AI-generated text, and ISBNdb pulled the sourcing page from its site this week.

One bookseller told 404 Media his weekly sales climbed from about 20 books to several hundred starting April. 

The orders had no theme, apart from every title carrying an ISBN. Dealers in the Netherlands, Germany, and Spain reported the same pattern.

"It benefits me financially as well as by clearing out old inventory that is otherwise unlikely to sell," one bookseller told 404 Media.

"On the other hand, I don't like the end-use, and I don't like that uncommon books are being pulped."

His inventory runs to rare and out-of-print titles, which means some of these orders are taking the last surviving copies.

Legal exposure keeps compounding for the buyers, which has turned clean provenance into the whole reason that print is worth the trouble.

This book-buying trend highlights several business implications:

• High-quality human writing has become a strategic asset. Companies should invest in original content to create resources that remain valuable as AI slop spreads.
• Legal sourcing is becoming a competitive advantage. AI developers should secure licensed or lawfully acquired content to reduce copyright risk.
• Content supply chains are changing. Publishers should evaluate licensing opportunities to create new revenue as demand for trusted training data grows.

As AI developers compete for cleaner datasets, the value of original content is rising.

Our Take: Is a Scanned Copy Still a Book?

AI companies are consuming the only input in their supply chain that nobody can produce more of.

We think that a scanned book stops being a real book once the paper hits the recycler, and no fair use ruling changes that.

The tell is the NDAs, since a company confident in its own defense would put its name on the invoice.

Publishers are getting played because the same first-sale rule that legalized this sends the resale money to dealers and none to rights holders.

They could be selling structured, licensed archives at a premium right now, with provenance guaranteed and nothing destroyed.

George Orwell put the memory hole inside a government ministry, but the real one turned up in a warehouse with a purchase order attached.

As AI accelerates the digitization of human knowledge, we should ask where convenience ends and cultural stewardship begins.

Looking for partners that can help your business adopt AI responsibly?

Explore these top AI consulting companies to find experts in governance, implementation, and enterprise AI strategy.