On August 3, Snopes confirmed what archivists had suspected for months: internal Anthropic documents, unsealed in a January lawsuit, show the company has been buying physical books by the millions — prioritizing “less common” titles — and running them through machines that “destructively scan” them under a program called Project Panama. The books do not survive the process. The pages are sliced, fed, and discarded.

The outrage this week has been loud and predictable. Anna’s Archive, the shadow library, issued a global call for volunteers to scan books before the AI companies get to them. “Time is running out,” the appeal says. Ten million volunteers, ten million books. The framing is familiar: knowledge is being monopolized on private servers, and the public must act.

But the framing misses the actual story. The reason AI companies are buying physical books is not that they want to hoard knowledge. It’s that the internet — the great digital commons that was supposed to make physical books obsolete — has become so saturated with machine-generated text that it is no longer usable as training data. The 404 Media report that broke this story quoted ISBNdb, a company that sources printed books for AI firms, telling clients that old books are valuable precisely because they are “guaranteed to be free of the very AI slop AI companies are producing.”

That sentence should stop everyone cold. The AI industry has polluted the digital world so thoroughly that the only clean data left is on paper. And now it is consuming the paper.

The Internet Is Now a Slop Field

For two decades, the open-internet crowd argued that digitization was preservation. Scan everything, upload everything, and the physical object becomes redundant. The Library of Alexandria burned once; the internet would never burn.

That argument is now dead. The internet did not burn. It filled with garbage. Every major training dataset is now contaminated with machine-generated text — blog posts written by chatbots, product reviews generated by models, Wikipedia edits smoothed over by LLMs. The 404 Media report put a number to the desperation: ISBNdb is marketing old printed books to AI companies explicitly because they are “free of AI slop.” Not because they are rare. Not because they are valuable. Because they are clean.

A rare-book dealer in Hudson, New York, who has watched his inventory get bought out by intermediaries over the past eighteen months, put it plainly: “They don’t want the book. They want the paper. The paper is the only thing left that hasn’t been through a model.”

Anna’s Archive Wants You to Save the Books — From the Wrong Villain

The irony of Anna’s Archive leading the preservation charge is almost too neat. This is a shadow library — the successor to Library Genesis and Z-Library — that has spent years fighting copyright lawsuits from publishers who argue that its entire existence is theft. Now it is asking the public to help it save books from destruction.

But notice the legal inversion. If an AI company buys a book and destroys it, that is legal. If a volunteer scans a book and uploads it to Anna’s Archive, that is a crime. The law protects the destroyer and punishes the preserver. The moral framing and the legal framing are exact opposites.

That inversion is not an accident. It is what happens when a society treats information as property first and culture second. The AI companies are not breaking the law. They are following it to its logical conclusion.

What Actually Gets Lost

The people who need to update their priors are not the AI executives. They know exactly what they are doing. The people who need to update are the open-internet utopians who spent twenty years insisting that the digital commons would outlast the physical one.

It did not. The digital commons is now a toxic waste dump, and the only pristine data left is on paper — which is exactly what is being destroyed. The tragedy is not that Anthropic is burning books. The tragedy is that the internet, which was supposed to be the great archive, has become so polluted that the only clean copy left is the one that gets sliced and fed to a scanner.

The books were the backup. Now the backup is being eaten.

Sources