Booksellers push back as AI firms seek books to scan for training

In short: More booksellers are getting wary of large, unusual book orders as reports say some AI companies buy and sometimes destroy books to scan them for training.

What's going on

Reports and social media posts are drawing attention to how some AI companies gather text from books. One fast method is to cut off a book’s spine, scan the pages quickly, and then throw the paper away. People worry this can lead to permanent loss, especially if rare books are involved.

Ars Technica points to an alternative. Google patented a non-destructive scanning method in 2009 that can scan books without taking them apart, although it can still create errors, like curved pages that bend the text or pages that get missed if workers move too fast.

The Internet Archive, a nonprofit that helps libraries preserve books, says it still scans books “the hard way,” one page at a time. A longtime scanner, Eliza Zhang, described using clean, dry hands to turn very thin pages, using a foot pedal to lift scanner glass, and double-checking fold-outs and blurry images. The goal is to avoid damaging books and to reduce mistakes.

What to watch

Booksellers are starting to report suspicious bulk purchases, which they say do not look like normal customer behavior. An Irish bookstore, Kennys, flagged an order for 5,000 obscure titles with no attempt to negotiate on price, which it called a red flag.

After a 2025 lawsuit revealed Anthropic destroyed millions of print books, pressure is growing on AI firms to show how they collect book text and whether they protect rare copies. Some companies say they will preserve rare books, but booksellers and readers are watching closely to see if those promises match what happens in practice.

Source: Arstechnica