AI companies are reportedly buying up and destroying millions of old books for training

Q: What is 'destructive scanning' and how are AI companies using it to digitize books?

Destructive scanning is a process where physical books are dismantled, spines removed and individual pages separated, so they can be digitized more quickly, with the source material effectively destroyed. AI companies are reportedly buying large quantities of books (including rare and out-of-print copies), having them scanned in this way to create digital training data, and sometimes working through intermediaries that source and supply books at scale under confidentiality agreements. Courts have treated such scanning as fair use in at least one case (Anthropic's Project Panama), and companies prize these human-written books as high-quality, untampered training data.

Q: How did the Anthropic 'Project Panama' court case influence the use of destructive scanning?

The Project Panama court case made destructive scanning widely visible by centering on Anthropic’s practice of buying and digitizing books then destroying them. The court ruled in Anthropic’s favor because it legally purchased the books for digital copying, but the decision prompted widespread backlash over the destruction of millions of books. As a result, AI companies have sought to shield themselves from criticism by working with intermediaries to acquire books at scale rather than directly sourcing them.

Q: Why are AI companies prioritizing older, human-written books described as free from 'AI slop'?

Because human-written books are seen as high-quality, untarnished training data in a post-generative AI world. The primary article says older books "free from any AI influence" are valued as training material precisely because they are "untainted by 'AI slop,'" meaning they are entirely written by humans and therefore presumed better for building and training new AI models.

Q: How are intermediaries protecting client identities when selling books to AI companies?

Intermediaries are promising confidentiality when sourcing large quantities of books and using strict nondisclosure agreements to shield their AI clients from exposure. The primary article cites ISBNdb as an example, which advertised sourcing books at scale with a "strict NDA" to protect AI companies.

AI companies are liaising with third parties to buy up massive quantities of books and then destroying them to train new AI models.

VIEW GALLERY - 3

Kosta Andreadis

Senior Editor

Published Jul 30, 2026 2:27 AM CDT

1 minute & 30 seconds read time

TL;DR: A 404 Media report says AI firms use "destructive scanning"-removing spines and pages-to digitize and destroy millions of books for model training, sometimes via intermediaries like ISBNdb that sourced large lots under NDAs; courts deemed such digitization fair use, driving demand for human-written books as high-quality data.

Voice: Kosta AndreadisSpeed

0:00 / 2:40

According to a new report from 404 Media, AI companies are not only using books to build and train powerful AI models, but they're also destroying them. Yes, to digitize millions of books in a way that saves them time and money, AI companies are removing spines and individual pages to make the whole process faster.

AI companies are reportedly buying up and destroying millions of old books for training 2

VIEW GALLERY - 3 IMAGES

It's a system referred to as 'destructive scanning,' where the source material is effectively destroyed. This process came to light a year ago, when Anthropic's "Project Panama" initiative was taken to court in the United States for this very reason. Even though the court ruled in favor of Anthropic because it legally purchased these books to turn them into digital copies, it led to widespread backlash because it meant that millions of books were destroyed, everything from paperbacks to rare prints and editions.

Even though destructive scanning is viewed as 'fair use' by federal courts, AI companies are now looking to shield themselves from criticism by working with intermediaries to acquire books at scale on their behalf.

The 404 Media report notes that these intermediaries promise confidentiality as they source physical copies of books in vast quantities, in the range of hundreds of thousands, or more, at a time. One of these companies is ISBNdb, which boasted that it could source rare books from "library shelves, used bookstores, and out-of-print catalogs," with the option to buy 1 million books at a time with a "strict NDA" to protect AI companies from being exposed.

AI companies are reportedly buying up and destroying millions of old books for training 1

That's 'boasted' in the past tense because this has since been taken down, in a move to minimize the fact that it's now making a lot of money from its "print books for AI training" business. But not before it was archived.

Frequently Asked Questions

TweakBot answers common questions about this news using TweakTown's own coverage from this page and related content from our archive. Tap a question to reveal the answer, or type your own below.

Question #1

What is 'destructive scanning' and how are AI companies using it to digitize books?

Question #2

How did the Anthropic 'Project Panama' court case influence the use of destructive scanning?

Question #3

Why are AI companies prioritizing older, human-written books described as free from 'AI slop'?

Question #4

How are intermediaries protecting client identities when selling books to AI companies?

Have a question not listed here? Ask below and TweakBot will answer it.

Based on the revelations, it certainly sounds like AI companies are trying to buy up every single book, digitizing them, and then destroying them. This isn't alarmist, but reality. And the reason it's happening, and reportedly accelerating, is that we're living in a post-generative AI world. The value of human-created works, especially in the literary space, is increasing. Untainted by "AI slop," these books that are free from any AI influence are viewed as extremely high-quality training data because they're entirely written by humans.