//

Court records show AI firms buying, destroying millions of books to train chatbots

AI developers are buying used and out-of-print books in bulk, scanning the pages into training datasets and destroying the physical copies, according to unsealed court filings and reporting on secondhand booksellers across Europe and the United States.

3 mins read
Vintage books in Zack Benz's collection, passed down through generations of his family. (Photo by Zack Benz)

Booksellers in the Netherlands, Germany, Switzerland and Spain say they’ve seen an unusual spike in bulk orders for obscure and out-of-print titles over the past year, and many suspect the buyers are artificial intelligence (AI) companies stocking up on training data.

A book database service called ISBNdb now facilitates these bulk orders, helping AI clients purchase anywhere from 1,000 to 1 million books at a time while keeping the buyers’ identities anonymous, according to reporting from 404 Media, which first detailed the practice.

“The world’s best AI training data is sitting on a shelf,” ISBNdb states on its website, as reported by 404 Media.

ISBNdb markets pre-2022 print books to AI clients because they predate the flood of AI-generated text now circulating online. The company’s own site states that “the world’s best AI training data is sitting on a shelf,” according to 404 Media‘s reporting cited by the Dallas Express. The company is said to employ both non-destructive scanning, using overhead or flatbed scanners, and destructive scanning, in which staff cut books apart and feed individual pages through high-speed industrial scanners. The destructive method is reportedly cheaper and faster, which is why AI firms favor it.

According to various court filings, AI companies are under scrutiny over how they acquire books to train their models, with allegations that they use copyrighted materials from unauthorized sources, such as shadow libraries. Companies such as OpenAI, Anthropic, Meta, Google, Microsoft and xAI have used millions of copyrighted books and are accused of copyright infringement by authors and publishers, while they argue that their practices qualify as fair use. This issue is now central to ongoing lawsuits and varying court rulings, with no clear legal consensus.

Vintage and modern books in Zack Benz’s collection. Most novels were passed down through generations of the Benz family. (Photo by Zack Benz)

Major AI providers are anonymously sourcing physical books through third-party middleman services, and secondhand booksellers say they cannot confirm which specific companies are behind the surge in orders

Court documents from the copyright lawsuit Bartz v. Anthropic describe how Anthropic used a hydraulic cutting machine to remove pages from books it acquired through resellers, then scanned the pages using industrial imaging equipment, as detailed by Futurism. The unsealed filings refer to the effort internally as “Project Panama.”

Anthropic’s practice relied on the first-sale doctrine, the legal principle that lets a buyer do what they want with a book once they’ve purchased it, without needing the copyright holder’s permission. In the Bartz case, U.S. District Judge William Alsup of the Northern District of California ruled last summer that scanning legally purchased physical books into digital copies, even when the originals were destroyed, counted as transformative fair use. Judge Alsup found the process qualified because the physical book was effectively replaced by its digital copy rather than duplicated for redistribution.

Vintage and modern books in Zack Benz’s collection. Most novels were passed down through generations of the Benz family. (Photo by Zack Benz)

That same Anthropic case separately ended with the company settling claims tied to pirated digital books, a distinct part of the lawsuit, for $1.5 billion. The fair-use ruling applies specifically to the print books Anthropic purchased and scanned itself, not the pirated material covered by the settlement.

Federal judges have since issued similar fair-use rulings in separate copyright cases involving OpenAI and Meta, though those rulings addressed training on written material generally rather than confirming the same buy-scan-destroy pipeline used in Project Panama.

The reasoning in Alsup’s ruling has drawn criticism from people in the book trade who argue it creates a financial incentive to destroy books rather than preserve them, since storing physical copies costs money while claiming the physical book was “replaced” by a digital scan strengthens a company’s fair-use argument in court. Critics have also pointed to an alternative approach: Harvard University’s partnership with Google and Microsoft to digitize public-domain books without destroying the originals.

Vintage books in Zack Benz’s collection, passed down through generations of his family. (Photo by Zack Benz)

Booksellers who have taken the orders describe mixed feelings about the situation. One anonymous bookseller told 404 Media the bulk sales help financially and clear out slow-moving inventory, but acknowledged uncertainty about what happens to the books afterward.

No government agency, library association, or industry group has issued formal guidance in response to the reports, and it remains unclear how many books total have been destroyed industrywide, since most buyers operate through anonymous intermediaries protected by non-disclosure agreements.

The Daily Planet will reach out to local sellers in the Twin Cities for responses in the coming weeks.

Zack Benz

Zack Benz has been a fan of the Daily Planet since he was eight years old. The Daily Planet has always been a beacon of hope for him and it’s his life’s mission to make it shine in a similar light to so many around the world. Zack graduated with a degree in journalism and art from the University of Minnesota Duluth in 2019.

Leave a Reply

Your email address will not be published.

Previous Story

Elon Musk’s xAI sues Minnesota over AI nudification law

0 £0.00