Rare books shredded to feed AI: Australian sellers sound the alarm

Key points

• Australian rare and secondhand booksellers believe old books are being purchased in bulk, scanned to create AI training data, then physically destroyed.
• Dealers say the books lost include rare and out-of-print titles that cannot be replaced once gone.
• The concern centres on AI training pipelines, the large-scale processes companies use to feed text into AI systems before the models can generate language.
• Melbourne bookseller Tim White, who runs Books for Cooks, is among those raising the alarm ahead of the Melbourne Rare Book Fair.
• No single company has been publicly identified as responsible for the buying and scanning activity.

Tim White handles a dust-jacketed book the way a surgeon handles an instrument: carefully, deliberately, with obvious respect. Setting up his stall at Melbourne University's Wilson Hall before the Melbourne Rare Book Fair, he put it plainly: "Sometimes books tell you more than the content."

For White and other Australian dealers in rare and secondhand books, the physical object matters as much as the words inside it. A handwritten dedication, a water stain, a particular edition's typeface: all of it carries meaning. Which is why what they believe is happening to some of their stock feels, to them, like destruction for destruction's sake.

What are sellers actually claiming?

Dealers believe buyers are acquiring old books in bulk, scanning them to create text datasets, then discarding or shredding the originals. Those datasets would feed AI training pipelines, the industrial-scale processes that technology companies use to teach large language models (AI systems, like the ones behind ChatGPT, that learn to generate human-sounding text by reading enormous quantities of existing writing) how to produce language.

The Guardian reported on the concern raised by sellers at the fair. No individual company has been named publicly, and no confirmed proof has emerged yet linking specific bulk purchases to a named AI firm. What sellers describe is a pattern: unusual bulk buying, sometimes of eclectic or out-of-print titles, followed by no trace of those books resurfacing in the secondhand market.

Why does this matter beyond book lovers?

Rare and out-of-print books are, by definition, finite. A first edition pamphlet or a locally printed title from the 1920s exists in maybe a handful of copies worldwide. Scan it and shred it, and the physical record is gone. Libraries and archives do digitise books, but they preserve the originals.

There is also a legal question sitting just below the surface. Authors and publishers have already launched lawsuits in the United States arguing that scanning copyrighted books to train AI without permission or payment violates copyright law. The physical destruction angle adds a further layer of concern, though Australian copyright law and its intersection with AI training remains largely untested in court.

For ordinary people, the practical upshot is this: the secondhand book market, already a fragile ecosystem of small independent dealers, may be losing irreplaceable stock to a supply chain most buyers know nothing about.

What should readers watch for?

If you sell books through online platforms or to bulk buyers you do not recognise, be cautious about unusually large or eclectic purchase offers, especially where the buyer shows no apparent interest in the titles beyond acquiring volume. Ask questions. Independent booksellers suggest dealing with known buyers and established fairs where provenance and care for physical objects are part of the culture.

Researchers and journalists tracking AI training data sourcing are the ones most likely to establish firmer links between bulk buying and specific AI companies. That work is ongoing.