What Happened
Tim White was setting up his stall at the Melbourne Rare Book Fair, lining up dust-jacketed books and pamphlets in plastic sleeves on temporary shelves inside Melbourne University's Wilson Hall, when the Guardian caught up with him. White runs Books for Cooks, an independent shop focused on the intersection of food and society. For him, a book's physical form carries meaning beyond its text.
"Sometimes books tell you more than the content," he said. "We're interested in where food and society connect. So we just keep pushing the boundaries, looking for things that show unusual aspects of that, things that tell us a bit more about what it is to be human."
That philosophy is now running headlong into the data hunger of generative AI. Vendors and rare book experts are increasingly worried that physical books, including one-of-a-kind artefacts, are being purchased specifically to be scanned and destroyed.
One seller described receiving a bulk order that felt off. The buyer paid in advance, raised no objection to postage costs, and ordered somewhere between 30 and 40 books. "It would have been about 30 to 40 books, and so finding them, packing them, making sure you hadn't missed one out. It was just driving me nuts," the seller said. Among the titles that went out was a poetry book that had sat on the shelf for two decades and sold for $9.
Why It Matters
The concern is not hypothetical. Earlier this year, unsealed court documents from a US copyright lawsuit brought against Anthropic by book authors revealed the company had been engaged in what the filings described as "destructive scanning": buying physical books, slicing off their spines to feed pages through scanners more efficiently, then pulping what remained. Anthropic had sought to keep the practice quiet. The presiding judge found it lawful under US copyright law, ruling it amounted to "transformative use."
That ruling applies only in the United States. Australian copyright law is distinct, and no equivalent ruling has been handed down here. But the legal question is almost secondary for sellers whose stock includes items that cannot be replaced.
"If it's a $20 book of which there's 500,000 copies in Australia, I wouldn't cry," one seller said. "I'd have different issues around copyright. If it's a book that has that storytelling element to it, or it's a one-off, it's horrific."
Key Details
White's framing of books as objects with stories of their own sits at the heart of what rare booksellers say is being lost. "That's what most of us [rare booksellers] do. We handle things that have a story," he said.
The buyers raising suspicion tend to share a profile: they pay upfront, they do not haggle, and they order in bulk across categories that would otherwise attract little commercial interest. Sellers have begun comparing notes at fairs and in trade circles, though no Australian investigation has been formally launched.
The Guardian's reporting, by Stephanie Convery, does not name the specific companies suspected of purchasing books for scanning in Australia. The factual record established in the US litigation centres on Anthropic, whose destructive scanning practice was confirmed through the unsealed court documents.
Background and Context
Generative AI models require enormous volumes of text to train on. Publicly available web text has limits, and publishers have largely refused to licence their catalogues at the scale or price AI companies want. Physical books, particularly older or out-of-print titles, represent a body of text that is harder to access digitally and has historically attracted less legal scrutiny.
Destructive scanning as a practice predates the current AI boom. Google's book-scanning project, which ran into its own copyright battles from the mid-2000s onward, used non-destructive methods. The shift toward spine-cutting reflects a drive for speed and volume rather than preservation.
For Australian sellers, the Melbourne Rare Book Fair represents a concentrated point of vulnerability. Titles that surface there once may not surface again.
What Comes Next
Australian copyright law does not contain a broad "transformative use" defence equivalent to the one applied in the Anthropic case. The Australian Law Reform Commission has previously examined fair use questions, and the current AI training debate may bring those recommendations back into focus for policymakers.
The Australian rare book trade has no formal mechanism yet for flagging suspicious bulk purchases or coordinating with regulators. Whether bodies such as the Australian Booksellers Association or the Australian Society of Authors move to address the issue remains to be seen.
White, for his part, keeps preparing his stall. The books go out on the shelves. Whether they stay in the world as objects, or end up as tokens in a training set, is a question the trade is only beginning to reckon with.