Australian booksellers warn AI book sourcing could put rare physical works at risk

Australian secondhand and rare-book sellers are raising a new concern about the generative AI supply chain: physical books may be bought in bulk for scanning, then destroyed after their text is digitized. The Guardian reported the concern from vendors at the 2026 Melbourne Rare Book Fair, tying it to broader industry anxiety after US court filings showed that destructive scanning has been used to build training corpora.

The story matters because it shifts part of the AI training debate from licensing and scraped websites to the fate of physical cultural material. Booksellers told the outlet that unusual bulk orders have arrived through online marketplaces, sometimes for obscure, low-priced or long-unsold titles. That pattern does not prove the books were destined for model training, but it has made sellers more cautious about where scarce or historically valuable items go after a sale.

What is known

• The Guardian report was first published on August 1, 2026, and updated on August 2.
• Several Australian vendors said they had received sizable or unusual orders from Zoom Books, a Canadian company that describes itself as a book recycler.
• Zoom Books told the Guardian that it buys secondhand books and resells them intact, and said it does not digitize books.
• An Anthropic spokesperson told the outlet the company had never bought from Zoom Books, while acknowledging that sourcing books is a common approach in language-model training.

The careful distinction is important. The report documents booksellers' concern and specific purchasing patterns, not a confirmed pipeline from those Australian shops to any named AI lab. It also includes denials from the two companies most directly discussed. Still, the concern is credible enough to highlight a practical gap in today's AI data market: once a physical book is sold through a general marketplace, a seller may have little visibility into whether it is preserved, resold, scanned, or destroyed.

For common paperbacks, that may be a commercial or copyright argument. For rare books, annotated copies, local histories, out-of-print technical manuals, or documents with provenance value, the loss can be more than a missing copy of text. The physical object may carry marginalia, binding details, ownership marks, inserts, or other historical signals that scanning cannot fully preserve.

The result is a small but telling pressure point in the AI economy. As model developers seek legally acquired text, intermediaries and secondhand markets may become more important. Booksellers now face a balancing act between moving inventory and protecting items whose cultural value is not captured by a price tag or a digitized page.

Sources

• www.theguardian.com

Cover photo by Mia's Photography on Pexels, used under the Pexels License.