Following the generative AI boom of 2022 and 2023, the race to build smarter chatbots became a race for data. Every major artificial intelligence company, including OpenAI, Meta, Microsoft, and Google, needed enormous amounts of high-quality human-written content to train and advance their models.

As publishers such as the BBC, the Daily Mail and other major news organisations began blocking AI crawlers, and copyright lawsuits against AI companies continued to mount, another challenge emerged. The internet itself was becoming saturated with AI-generated text, making it harder to separate original human writing from machine-generated content.

As the availability of training data shrank, books, particularly those published before 2022, when AI-slop had not yet begun to flood the web, became one of the last large collections of untouched human knowledge.

The Secret Project Panama
That strategy has landed AI company Anthropic, the creator of Claude, in trouble after court documents revealed details of a secretive project known internally as “Project Panama.” According to the federal court filings, the company set out to acquire as many books as possible, with internal documents stating that the goal was to “destructively scan all the books in the world” while keeping the effort out of the public eye because the company executives expected backlash.

Having purchased millions of physical books, Anthropic chose a faster and cheaper alternative to scanning them by hand. The company is said to have used industrial equipment to slice off the spines, allowing the loose pages to pass through high-speed scanners. Once digitised, the originals were discarded, permanently destroying the physical copies. The approach accelerated the creation of a digital library for AI training, but at the cost of preserving the books themselves.

What worries many is that there appears to have been no systematic effort to determine whether some of those books were rare, out of print, or among the last surviving physical copies. Publishers, writers, and literary advocates argue that books that had survived wars, fires, floods and decades of handling may have been permanently removed from circulation to become training data for an AI model. Critics online, including tech billionaire Elon Musk behind xAI, argue that while the text now exists in a private digital archive, the physical cultural artefacts should be preserved rather than lost forever.

“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,” Musk wrote on X.

Anthropic Pirating Books
The controversy extends beyond the destruction of physical books. Anthropic also faced allegations that it used millions of pirated books downloaded from shadow libraries such as Library Genesis (LibGen) and Pirate Library Mirror before later purchasing physical books for scanning.

Anthropic $1.5 Billion Settlement
While a federal court ruled that scanning legally purchased books and using them to train AI qualifies as fair use under US copyright law, the separate issue of acquiring pirated books resulted in a $1.5 billion class-action settlement involving more than 500,000 authors and publishers. As part of the agreement, Anthropic did not admit wrongdoing.

The case has poured fuel on an already heated debate over how AI companies obtain training data. Supporters argue that buying books gives companies the legal right to digitise them and that AI training transforms the material into something new. Critics counter that destroying millions of books, especially rare and hard-to-find titles, creates an irreversible cultural loss while concentrating humanity’s knowledge inside private AI systems that the public cannot access.

Anthropic Double Standards
The irony is that, in recent months, Anthropic has publicly warned competitors in China aginst copying their Claude AI capabilities and intellectual property. For many observers, this sits uneasily alongside revelations that the company developed part of its AI training library by scanning millions of books and, most controversially, destroying the physical copies afterwards.

Regardless of whether it was legal or moral, Anthropic’s practice has raised difficult questions about copyright, preservation and the cost of building the next generation of artificial intelligence.