AI Is Destroying Books to Save Itself

Walk into the warehouse and the first thing you hear is machinery.

Scanners humming. Pages turning. Hydraulic cutters slicing the spines from physical books.

Millions of books sit on shelves or in stacks across the floor, waiting to be processed. Each book is taken apart, scanned page by page and eventually sent away for recycling.

It sounds like a scene from Fahrenheit 451.

Except this warehouse is real.

According to internal documents revealed through litigation, Anthropic—the artificial-intelligence company behind Claude—created a secret initiative known as Project Panama.

Its ambition was stated plainly:

“Project Panama is our effort to destructively scan all the books in the world.”

The same internal document reportedly warned employees that Anthropic did not want the project publicly known.

Why would an AI company destroy millions of books?

Because the most advanced AI systems in the world are starving for something increasingly difficult to find:

Verified human knowledge.

Anthropic purchased physical books, removed their bindings with industrial cutters and ran the loose pages through high-speed scanners. The resulting digital copies became part of an internal research library. The physical books were not preserved.

Some vendor proposals reportedly discussed processing as many as two million books within six months.

Anthropic even hired Tom Turvey, a former Google executive who helped negotiate partnerships for Google Books, to lead the effort.

This was not a side project.

It was an industrial-scale operation to turn physical human knowledge into machine-readable data.

The destruction was part of the legal strategy

The strangest part of the story is that destroying the books strengthened Anthropic’s legal position.

In 2025, federal Judge William Alsup ruled that converting lawfully purchased print books into digital copies for internal use constituted fair use. Because each purchased book was destroyed during the scanning process, the digital copy replaced the physical copy instead of creating an additional copy for distribution.

One copy went in.

One copy came out.

The court viewed that process as a permissible change in format rather than the expansion of Anthropic’s library.

The ruling did not mean that every book Anthropic possessed was legally obtained. The company faced separate claims involving hundreds of thousands of books downloaded from pirate libraries.

That distinction proved enormously expensive.

In July 2026, a federal judge granted final approval to a $1.5 billion settlement covering 482,460 books allegedly acquired through piracy. Authors and publishers are expected to receive approximately $3,000 for each eligible work. The settlement addressed how those books were obtained—not the broader question of whether AI training itself qualifies as fair use. AP

Anthropic learned that it could not safely pirate the books.

So it purchased physical copies and destroyed them instead.

That may be legal under the court’s interpretation. Whether it is ethical is a different question.

The internet is becoming contaminated

The larger story is not about scanners or warehouses.

It is about data quality.

Since the explosion of generative AI after 2022, the internet has been flooded with machine-generated articles, summaries, product descriptions, comments, images and social posts.

AI companies increasingly face the possibility of training their new models on content created by older AI models.

That creates a dangerous feedback loop.

A model generates content. That content enters the internet. Another model trains on it, reproduces its errors and adds new ones. Over time, synthetic material can overwhelm the original human work.

The internet begins feeding AI back to itself.

This is why books published before the generative-AI boom are becoming premium inventory. Not necessarily because they are rare, but because companies can be reasonably confident that a human wrote them.

Those books represent uncontaminated intellectual material:

Original ideas.

Human experiences.

Carefully developed arguments.

Expert knowledge.

Distinctive voices.

They are records of human thought produced before machines began imitating human thought at scale.

Human data could become more valuable than compute

For years, the AI race has been framed primarily as a competition for computing power.

Who has the most advanced chips?

Who can build the largest data centers?

Who can afford to train the biggest models?

Compute will remain important, but technology typically becomes cheaper and more accessible over time. Chips improve. Infrastructure expands. Training becomes more efficient.

Verified human data does not follow the same curve.

The supply is limited.

Every time the internet becomes more saturated with synthetic content, authentic human work becomes harder to identify—and therefore more valuable.

That changes the economics of creativity.

Writers, filmmakers, athletes, coaches, researchers, teachers and subject-matter experts should stop thinking of their work as “just content.”

They are producing proprietary human data.

A veteran basketball player explaining how he reads a defense is creating something a machine cannot independently experience.

A coach documenting 30 years of pattern recognition is building a knowledge asset.

An entrepreneur recording the mistakes behind a successful company is producing intelligence that cannot be recreated by scraping generic business articles.

AI can summarize this knowledge.

It can reorganize it.

It can imitate its language.

But first, a human being has to live it, understand it and put it into the world.

The opportunity for creators

Project Panama sends a powerful signal.

One of the world’s leading AI companies was willing to operate warehouses, purchase millions of books, hire specialized personnel, cut the books apart and build an enormous private digital library.

That is how valuable authentic human knowledge has become.

Creators should pay attention.

Your archive matters.

Your interviews matter.

Your private research matters.

Your newsletters, books, courses, recordings and databases matter.

Most importantly, ownership matters.

If you publish everything exclusively on social platforms, you are helping build somebody else’s data library. You may receive attention, but the platform retains the audience, controls the distribution and captures much of the long-term value.

This is another reason creators need their own websites, email lists, archives and intellectual-property systems.

Social media is rented land. Human knowledge is owned property.

The coming premium on authenticity

Project Panama looks like a story about the destruction of books.

It is actually a story about the rising value of humanity.

AI can generate unlimited words, but unlimited supply does not create unlimited value.

The scarcer asset is becoming proof that a real person created something through real experience, original research and independent thought.

In the next phase of the AI economy, the most valuable label may not be “AI-powered.”

It may be:

Human-made. Human-verified. Human-owned.

Compute will eventually become cheaper.

Authentic human intelligence will not.

Human Knowledge Is Becoming the New Scarce Asset

AI can generate unlimited content—but it cannot manufacture lived experience, original thought or authentic human perspective.

That’s where your value lives.

If you believe creators should own their ideas, control their audiences and build assets on land they actually own, subscribe to Athletic Entrepreneur.

Then share this story with someone creating valuable human work before the machines absorb it.

Subscribe. Share.