When old knowledge becomes new: How AI companies destroy books and hoard knowledge

Anthropic and Amazon buy millions of books to train their AI models Claude and Nova.

In this process, the often rare editions are cut up and then shredded.

The monopolization of freely accessible knowledge is therefore already in full swing. – A comment from Mikhailo Böhnisch.

In January of this year, court documents from the case Bratz vs.

Anthropic PBC” released in 2024.

They revealed that the Google-backed AI company had purchased and destroyed millions of printed books.

In the effort called “Project Panama,” the books were “destructively scanned.”

Spines and covers were removed using hydraulic cutting presses and the individual pages were fed into high-speed scanners.

The originals were then shredded and sent to landfills.

The “Project Panama” has been running for several years and has received support from Google’s ranks.

In February 2024, Anthropic hired the former head of the partnership program for Google's book scanning project, Tom Turvey.

Above all, he should ensure that the project does not encounter “ legal, practical or business obstacles ”.

Obstacles such as copyright.

Anthropic cleverly ignored this until the class action lawsuit in 2024 by feeding its AI language model Claude with pirated books.

Billion dollar lawsuit against Meta: Profit takes precedence over child and youth protection Loopholes in copyright Antropic is said to have illegally downloaded and stored seven million books as part of the project.

The aim was supposedly to create a “ central library ”.

However, Anthropic's self-serving intentions are loud and clear from internal documents.

For example, from Exhibit 21 in the court file.

In it, Tom Turvey explains: “Project Panama is our effort to destructively scan all of the world’s books.” The internal memo also encourages Antropic workers to exercise discretion, “Because we don’t want it to be known that we’re working on this.” This desired clandestineness ultimately had financial consequences. The company had to pay $1.5 billion to reach an out-of-court settlement in the copyright dispute.

With sales of $14 billion (as of 2025), there can hardly be any talk of a loss for Anthropic.

Rather, Judge William Alsup's ruling creates another loophole.

The judge ruled that using copyrighted material to “train” an AI language model does not constitute copyright infringement. “This is not the type of competitive or creative exclusion contemplated by copyright law,” he wrote in his ruling .

He also maintained that, as part of this process, the printed copies were replaced by digital versions.

Scanning and feeding into AI models would not create new copies or distribute them commercially.

The procedure is therefore legitimate in terms of copyright law.

Europe is far behind in the AI ​​arms race Accumulation and destruction The fundamental necessities and contradictions of capitalism do not stop at large AI companies.

To generate a profitable product and maintain capital accumulation, Anthropics Claude must be “trained” with a high-quality voice dataset.

Preferably with long texts that were created before 2002.

ChatGPT came onto the market in 2002 and has had an influence on contemporary texts ever since.

To improve the ability to predict linguistic outcomes, an AI model needs as many language combinations as possible.

In large quantities and high quality.

A quality that has no longer existed since 2024 at the latest.

Generative AI models like Claude or ChatGPT are starting to show signs of model collapse.

This collapse occurs when a model is “trained” on synthetic data.

So data that was generated by other AI models.

According to court documents, Anthropic hoped that the books they misused would "help Claude write as accurately and compellingly as authors do."

Amazon also destroys books But Anthropic is not the only one who uses “destructive scanning”.

On the 17th

In August 2026, 404 Media published a research article.

It revealed that the delivery of nearly 1,000 rare books ended up at an Amazon AI training center.

ISBNdb, another focus of the 404 article, touts itself as the world's largest book database.

But most importantly, the site helps AI companies buy between 1,000 and a million books per order.

As a bonus, there is another promise: the companies' purchases will be kept secret by ISBNdb.

After all, we want to prevent a public outcry like what happened with Anthropic.

Everyone is aware that this approach can seem shady.

ISBNdb itself writes on its own website (now archived): "The image problem is real. 'AI company destroys two million books' is not a headline that arouses sympathy." The end of the “liberal” Internet?

Google's AI as a predator and censorship machine Cultural barbarism The destruction of knowledge has been a favored means of oppression since ancient times.

Where the erasure of oral traditions in the course of subjugation and genocide ends, the systematic destruction of written knowledge begins.

In the 1560s, imperial Spain destroyed the writings of Yucatán in the struggle for colonial supremacy in Latin America.

Even during the slaves' struggles for self-emancipation, writings about the abolition of slavery were repeatedly burned in the southern states of the USA.

Written knowledge, especially publicly available or progressive, has always been the focus of destruction by authoritarian forces.

Because if this knowledge disappears, the public perception of general society can easily be controlled and enforced.

The vacuum created by the annihilation can be filled with one's own ideological values.

The looting of the German Institute for Sexual Sciences on January 6 shows what an immense vacuum can be created.

May 1933 by the Nazi regime.

From research into the demonization of homosexuality to the scientific foundation of trans identity as a natural phenomenon.

The research and findings of Magnus Hirschfeld and his colleagues at the institute were revolutionary.

Nothing of this accumulated knowledge, such as the first documented gender reassignment operations, remains today.

Only a few of Hirschfeld's personal documents survived the systematic destruction.

The rest, and thus an enormous wealth of knowledge, is lost forever to us future generations.

AI should have a say in asylum and visa procedures “Where books are burned…” The implications for society are almost identical in the case of book destruction by Anthropic and Co.

Feeding knowledge from sometimes rare books to generative AI models supports the violent dominance of colonial-imperialist power blocs.

This appropriation becomes necessary within the framework of the capitalist logic of property.

In order to maintain the capitalists' monopolies on power, potential power from below must be actively prevented.

For a system that uses fascistization processes to maintain itself, nothing is as threatening as an educated working class.

But if it can be guaranteed that writing and reading skills in the working class are kept low, then the revolutionary potential will be nipped in the bud.

What would be better suited for this than a freely accessible program?

Better yet, if this program is prone to hallucinations and collapse or induces psychosis.

Instead of being the creators and preservers of our knowledge, we are converted into consumers.

The pursuit of ever larger AI models has exhausted the supply of human text on the Internet.

But in order to maintain their growth, companies have to generate profits and to do this, books have to be destroyed.

A logical conclusion of capitalist interests.

The supposed right to a monopoly on power is also easier to defend if knowledge is made a rare raw material.

But the “intelligence” that Anthropic and generative AI companies are trying to build is, at its core, a cannibalistic computer program.

It devours one's own past and accumulated human knowledge.

Just to be able to best guess the next sentence it should generate.

As with the recent flood disaster in Nepal and Tibet, the consequences for people and the environment are of no concern to the program or to companies.

The implications here are, as with any historical example of the destruction and hoarding of knowledge.

As early as 1823, the German poet Heinreich Heine declared: “Where you burn books, you also end up burning people.” The post When old knowledge becomes new: How AI companies destroy books and hoard knowledge appeared first on Perspective.