‘More than just objects’: Australian rare book sellers warn AI is destroying cultural treasures

Australian rare book sellers are sounding an urgent alarm over what they describe as the “horrific” destruction of irreplaceable titles, warning that a growing number of books are being pulled from circulation not to be read, but to be dismantled, scanned, and fed into artificial intelligence training pipelines. Dealers across Sydney, Melbourne, and Brisbane report that they have been approached in recent months by brokers and intermediaries acting on behalf of technology companies, seeking large quantities of rare, out-of-print, and collectible volumes. At first, these approaches looked like ordinary bulk purchases. On closer inspection, however, the buyers were not interested in preserving the books or reselling them — they wanted the pages stripped and digitized, and the physical remains discarded.

For bibliophiles, the practice is difficult to stomach. “These books are more than just objects,” said one veteran Melbourne dealer, who asked to remain anonymous to protect his business relationships. “They’re records of human thought, history, and craft. Once you cut the spine and run the pages through a scanner, you have lost something that can never be put back.” A scanning operation may take minutes, but the book being destroyed may have survived centuries — through wars, fires, floods, and shifting fashions.

Why AI needs to eat books

The incident sits at the intersection of two increasingly uncomfortable realities. The first is that modern large language models — the systems behind chatbots, writing assistants, and search tools — depend on staggeringly large collections of text to learn grammar, facts, reasoning, and style. The industry is now facing a shortage of high-quality training data, and researchers have begun to look beyond the vast, freely available text of the open internet to the far richer and more carefully edited contents of books.

The second reality is cold commercial logic. Books are highly structured, professionally edited, and semantically dense — everything that a machine learning engineer could want. A single rare book may contain tens of thousands of words as useful as a hundred blog posts. And out-of-print or orphaned works, whose copyright status is murky, make tempting targets precisely because they are difficult for rights holders to trace. Tech firms and their intermediaries have reportedly offered significant sums for large collections, giving struggling bookshops a difficult financial decision. But the darker underside of this trade, critics say, is that money has begun to trump care.

More than just data: destruction of material culture

For book historians, the alarm goes beyond the loss of the words. A printed book is an artifact. Its paper, typeface, binding, illustrations, margins, annotations, and even its smell and wear carry information about how it was made, sold, read, and preserved across the centuries. Marginalia, in particular, is often the only surviving record of how past readers understood a text. A high-resolution scan can reproduce the surface of a page, but it cannot easily convey the depth of the paper, the texture of the leather, the watermarks, or the three-dimensional physical reality of a codex that has been handled by dozens of generations.

“The book is the container, and when you destroy the container, the contents become ghosts,” said Dr. Eleanor Whitfield, a preservation specialist at a leading Australian university library, in an interview. “Digitization is wonderful for access. But it is not preservation of the physical object, and it is certainly not a substitute for the cultural value of the original.” Whitfield pointed to the fate of Victorian-era illustrated natural history volumes, many of which have already been digitized, yet whose hand-colored plates still command extraordinary prices because collectors and researchers prize the physical plates as works of art.

A market under pressure

Booksellers are now caught between the demands of a lucrative new class of buyer and their own professional ethics. The Australian and New Zealand Association of Antiquarian Booksellers (ANZAAB) has issued an informal warning to members, urging them to scrutinize bulk purchase inquiries for signs that books will be destroyed. Dealers say the pressure is real: with physical retail foot traffic down and online sales squeezed by global platforms, the offer of thousands of dollars for a shelf of unwanted stock is hard to refuse — particularly for siblings and heirs of deceased collectors who want the books gone quickly.

Some booksellers have responded by inserting clauses into sale contracts prohibiting the dismantling of books or requiring buyers to confirm that volumes will be kept intact. Others have begun quietly blacklisting brokers associated with book-breaking operations. But enforcement is difficult. Once a book changes hands, the seller has no practical way to track its fate. “We sold a collection of Australian poetry to a man who said he was a librarian,” recalled a Sydney dealer. “A month later, one of the books appeared on an image-sharing website with its spine clearly cut. There’s nothing we could do.”

The copyright thicket and the training-data race

The destruction also deepens an already thorny legal debate over copyright and AI training. In the United States, several class action lawsuits by authors have challenged the notion that mass ingesting of copyrighted books for machine learning is “fair use.” In Australia, copyright law allows some reproduction for research and study, but the situation for commercial AI training remains largely untested. Destroying physical copies while digitizing them only complicates the evidentiary picture: when an original is gone, proving ownership of a specific edition — and therefore asserting rights over its text or illustrations — becomes far more difficult.

Legal scholars say the practice exploits a gray zone. “If you burn a book after scanning it, you’ve destroyed the best evidence of what you copied,” said one intellectual property law professor who asked not to be named. “That’s not necessarily illegal, but it is troubling from an accountability perspective.” The professor noted that image models, which are trained on scanned illustrations and photographs, pose a parallel risk: the original art may be destroyed after capture, leaving no physical record for artists who wish to contest infringing outputs.

What happens next

So far, Australian institutions have been slow to respond. The federal government has not issued guidance on the destruction of books for AI training, and the debate in Canberra has centered on copyright liability rather than physical heritage. But the bookselling community is pressing for action. Some dealers are lobbying for inclusion of rare books and manuscripts under national heritage protections, which would restrict exports and, in some cases, prevent destructive uses. Others are calling on auction houses to disclose when purchases are made on behalf of technology companies, and on libraries to deposit at-risk materials in secure archives.

In the longer term, the AI industry may need to rethink its approach. Several AI developers have begun experimenting with synthetic data and licensing deals with libraries and newspapers as more ethical alternatives. Project Gutenberg, the Internet Archive, and HathiTrust offer millions of digitized books that are already out of copyright, though the quality and range of these collections are limited compared with the totality of published human knowledge. Some researchers suggest that the market for destroyed books will naturally shrink as these legal and licensed pipelines mature — but that offers cold comfort to the dealers and collectors who must watch their shelves empty at an ever-quickening pace.

The deeper question, according to Whitfield, is one of values. “Every culture decides what it is willing to destroy in order to build something new,” she said. “Artificial intelligence is a remarkable technology, but it is not worth the same price for every book. Some books are simply irreplaceable, and when we destroy them, we are not feeding intelligence — we are feeding our own forgetfulness.”