Rare Books at Risk as AI Training Drives 'Destructive Scanning' Trend

Australian secondhand booksellers are raising alarms over a surge in high-volume, price-insensitive orders for rare titles, sparking fears that physical books are being acquired for "destructive scanning." This industrial process involves slicing off book spines to scan pages more efficiently before pulping the remains to feed generative AI training datasets.

Several Australian vendors, including Sainsburys Books and Good Reading Secondhand Books, reported receiving bulk and random orders from Zoom Books, a Canadian firm that describes itself as a book recycler. While Zoom Books denies digitizing books and claims it resells them intact or recycles them responsibly, the pattern of acquisition has drawn scrutiny. The concern is amplified by unsealed US court files revealing that the AI company Anthropic has engaged in destructive scanning—buying books, slicing their spines, scanning the content, and pulping the physical copies. Anthropic has stated it does not buy or destroy rare or antiquarian books and has never purchased from Zoom Books.

The Hunger for Pre-AI Data

Generative AI companies require massive volumes of high-quality text to refine their models. As digital archives become saturated or "polluted" with AI-generated content—often referred to as "AI slop"—there is a growing demand for physical, pre-AI texts. This has created a lucrative market for intermediaries and recyclers who can source physical books in bulk for digitization.

This shift in sourcing has significant legal backing in the United States. In the case of Bartz v. Anthropic, a US judge ruled that destructive scanning for the purpose of AI training constitutes "transformative use" and is therefore lawful under US copyright law. This ruling effectively provides a legal shield for companies that destroy physical media to extract digital data.

A Threat to Cultural Heritage

For bibliophiles and historians, the practice is viewed as a cultural catastrophe. Rare books are not merely data carriers but historical artifacts; their destruction removes the provenance and physical context of the work. Tim White, owner of Books for Cooks, described the process as "horrific" for one-off titles or books with a storytelling element. Gwenyth Todd of Chatelaine Books echoed this sentiment, noting that books are "more than just objects."

By erasing unique copies of niche or historical texts, the industry risks permanently deleting the physical record of human knowledge in exchange for a digital approximation. The loss of these artifacts means that future scholars will lack access to the original materials, leaving only the data processed by a corporate algorithm.

What to Watch

As AI companies continue to exhaust digital sources, the pressure on the global secondhand book market is expected to increase. The primary point of contention remains the transparency of the supply chain between "recyclers" like Zoom Books and the AI labs that require the data. Whether other jurisdictions will follow the US lead in classifying the destruction of physical books as "transformative use" remains a critical question for the preservation of global cultural heritage.