Policy

AI Training Destroys Millions of Books, Bookseller Warns

Court-approved scanning practices let AI companies buy, scan, and shred books at scale—including rare titles that may never be replaced.

Omega Editorial· August 28, 2026· 3 min read

AI Training Destroys Millions of Books, Bookseller Warns

AI companies are destroying far more books than previously reported—potentially tens of millions—as they race to train large language models, according to a bookseller who has tracked the practice extensively.

Scott Brown, owner of Downtown Brown Books in Portland, Oregon, has documented the book-scanning ecosystem on his Substack, Dispatches from the Rare Book Trade. He warns that the scale of book destruction extends well beyond what court documents from Anthropic's recent lawsuit revealed.

"Everyone lost their minds over Project Panama, and that is so small in comparison to what's really happening today," Brown said in an interview with The Orange County Register.

How the Destruction Works

The practice became public through the class-action lawsuit Bartz v. Anthropic, which exposed the company's "Project Panama"—an initiative to "destructively scan all the books in the world." Court documents revealed contracts calling for up to 2 million books to be scanned and destroyed.

The destruction is intentional, not accidental. Under the judge's ruling in the case—finalized last month with Anthropic paying a $1.5 billion settlement—AI companies can copy a book's contents for training as long as they destroy the physical copy and recycle it. One book in, one book out.

Reporting by 404 Media traced a large book shipment to an Amazon warehouse in Las Vegas, where employees confirmed they receive bulk orders, cut off bindings, scan pages, and destroy the books. Brown notes that non-destructive scanning methods exist, as demonstrated by the Internet Archive, but AI companies aren't using them.

Why it matters

This practice threatens the availability of books that exist in limited quantities. While millions of copies of bestsellers remain in circulation, rare and nonmainstream titles—sometimes with only five or ten copies available worldwide—are being systematically removed from the market. Multiple AI companies need the same titles, meaning each book may be purchased and destroyed multiple times by different firms. The scanned content trains the AI models but isn't preserved in any accessible database, meaning the books are gone permanently.

Pressure on Booksellers

The practice creates difficult choices for independent booksellers. Canceling suspected AI orders damages their fulfillment ratings on marketplace platforms, potentially dropping them from five-star to three-star status and making them appear less reliable to human customers.

"Booksellers are just in a difficult situation because if we reject orders that come through these marketplaces, our fulfillment rate... gets dinged," Brown explained. He noted one colleague received a 51-book AI order yesterday—too many to cancel without severe platform penalties.

Brown himself has received five AI-related orders. He declined to send one book—a signed Ray Bradbury novel—and spent an hour finding another seller who could provide an unsigned copy instead.

Nathan Gabbard of Los Angeles's Owl Creek Books, which specializes in rare and vintage editions, said he hasn't knowingly sold to AI companies but has heard about the practice from other dealers. "If I'm entrusted as a rare antiquarian book dealer, as the custodian of these items while they're here, I want to find these items a good home where they will be appreciated," he said.

Brown doesn't expect the practice to stop but hopes AI companies will avoid purchasing unique or signed copies, focusing instead on common editions in poor condition. "In order to not destroy books, you have to go to a lot of work," he said. "There is no good solution for booksellers."

These details were first reported by Erik Pedersen for The Orange County Register.

#ai training data#book destruction#anthropic#rare books#booksellers#copyright

This is an original analysis by the Omega editorial team. Source reporting: AI Watch.

Want systems like this working for your business?

Book a Call

More in Policy

Policy· 3 min read

Climate Activists Launch 'Irreplaceable' to Fight AI Industry

Veterans of the Sunrise Movement and 350.org are applying protest tactics from climate campaigns to challenge tech companies and government AI policies.

Via AI Watch · Aug 28, 2026
Policy· 3 min read

CIA Uses GPT-4 to Generate Intelligence Reports for Police

Documents reveal the agency is feeding international news into ChatGPT and distributing AI-summarized analyses through DHS fusion centers.

Via AI Watch · Aug 28, 2026
Policy· 4 min read

Federal Court Rules Private AI-Generated CSAM May Be Protected Speech

Seventh Circuit decision highlights gap between child protection laws and generative AI capabilities, urging Supreme Court review.

Via AI Watch · Aug 28, 2026