Are AI companies scanning and destroying millions of books, including rare titles?

In July 2026, a rumor circulated online that artificial intelligence companies have been scanning millions of physical books, some of which are rare or obscure, and destroying them in the process. 

Social media users made the claim on platforms such as Instagram, Facebook and Reddit. One popular Instagram post included an image, featuring text that focuses on artificial intelligence company Anthropic. It reads (archived, detail in brackets ours):

Al companies like Anthropic are buying obscure and rare physical books in bulk, cutting them apart and scanning them for training data, secretly destroying millions of books to build Claude.

btw [by the way] anthropic’s internal document on this literally said “we don’t want it to be known that we are working on this.”

it was called project panama.”

Several news media outlets also made the claim. For example, Futurism, a publication dedicated to covering science and technology that directly impacts markets and society, published an article on July 25, 2026, with the headline: “AI Companies Are Buying Antique Books, Ingesting Their Contents to Train Models, and Then Destroying Them at Incredible Scale.”

Dozens of Snopes readers emailed us and searched the website, seeking to confirm the veracity of the rumor. 

The claim appears to stem from 404 Media, an independent, investigative, digital media company. On July 21, 2026 (archived), the outlet published an article that quoted ISBNdb, a company that describes itself as the world’s largest book database. The passage reads, “The optics problem is real. ‘AI company destroys two million books’ is not a headline that generates sympathy.” The quote could be found on a now-removed page on ISBNdb’s blog (archived) that said Anthropic had been acquiring physical books, scanning and destroying them. “Two million books. Bought, read by machines, and returned to pulp,” the page read.

(404 Media’s article says ISBNdb helped AI companies source books for purchase, but ISBNdb later updated its website, saying: “The facts: ISBNdb has never purchased, scanned, or sold a book – for AI training or anything else. We don’t train AI models, and we never have. The page was a test of market interest; no such service was ever brought to life. We’ve taken the page down.”)

A review of Anthropic’s internal documents, made public in January 2026 in the context of a lawsuit, revealed it is true that Anthropic has been acquiring physical books in bulk, prioritizing “less common” books, and “destructively scan[ning]” them as part of an effort the company called “Project Panama.” 

Snopes contacted Anthropic to ask if Project Panama is ongoing as of this writing. We will update this article if the company responds.

It was not possible to confirm whether other AI companies have acted in a similar fashion. Given that numerous social media users claimed that multiple AI companies had participated in destroying books to train their models, we have rated the rumor mostly true but with undetermined elements.

Anthropic trains its AI model on physical books

ISBNdb said in a separate blog post (now deleted, archived here) that AI companies seek to train their models on books as high-quality data to prevent “model collapse,” which happens when AI trains on its own content. This feedback loop degrades the quality of AI output and results in AI slop. 

In 2024, several authors filed a lawsuit against Anthropic, alleging the company trained its model on their work without permission. The lawsuit says this was copyright infringement. Anthropic defended itself, saying it was fair use. 

In the context of this lawsuit, the judge ordered Anthropic to unseal internal documents about an effort it called Project Panama. 

Led by Tom Turvey — who had previously worked at Google creating partnerships to develop, among other products, Google Books — Project Panama was meant to be kept secret. As an internal memo from April 2024 shows, its goal was to scan and destroy “all the books in the world”:

Project Panama

Owner: Tom Turvey

Last major update: Apr 13, 2024

What is Project Panama?

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

Why use a codename?

We use a “soft codename” for it because we don’t want it to be known that we are working on this. This document is visible to all Anthropic employees, but you should avoid talking about it in public areas, and the fact that we are working on this should not be shared with anyone outside Anthropic.

(In order to quickly scan books, they must be sliced at the spine to feed the pages to a scanner, according to 404 Media.)

Another 211-page document, partially redacted, describes the process. On Page 10, it explains how to categorize books. One example it uses is a link to the website of the Book Industry Study Group, a publishing industry group. That link led to the category of “Antiques and Collectibles.” It is not clear whether this means Anthropic was targeting antique books or books about antiques.

An email exchange commencing on Aug. 27, 2024, between Turvey and Wonder Book, a company that sells books in bulks, shows Turvey confirming that “less common books are a great place to start.” This confirmed the part of the claim that Anthropic was targeting rare books.

Anthropic reached out to several companies that sell books, including in Spanish, as this exchange with Spanish publisher Océano shows. (Océano also has an office in Mexico.)

An October 2024 internal memo (Page 2) shows the company had already purchased millions of books at that point in time. (Though the exact number is redacted, the M at the end of the number, which stands for million, reveals the magnitude of the project). 

For further reading, Snopes has investigated several claims about AI companies, including rumors about their impact on the environment.