‘More than just objects’: Australian booksellers raise alarm over ‘horrific’ destruction of rare titles to feed AI

{"title":"Australian Booksellers Decry 'Horrific' Destruction of Rare Books for AI Training","excerpt":"Australian rare-book dealers are raising urgent concerns over the wholesale destruction of irreplaceable titles, reportedly pulped and digitized to feed artificial-intelligence systems. Sellers warn that cultural heritage is being erased for the sake of algorithms, while industry groups call for new protections and transparency from tech companies.","content":"## A Quiet Crisis in Australian Bookshops\n\nAcross Australia, a disturbing practice has booksellers and archivists sounding the alarm: rare and out-of-print titles are being destroyed — not by fire, flood, or neglect, but by the quiet machinery of artificial-intelligence development. Dealers in first editions, local histories, and irreplaceable literary works say they are increasingly seeing evidence of books being purchased, stripped, scanned, and then discarded or pulped, their contents fed into large language models that underpin modern AI chatbots. The trade has condemned the practice as \u201chorrific,\u201d arguing that these volumes are \u201cmore than just objects\u201d — they are fragments of cultural memory that no algorithm can truly preserve.\n\nReports suggest the practice is not limited to Australia. Global technology firms and AI startups have been sourcing vast quantities of printed material to train their systems, and partially digitized or damaged books are often considered low value once scanned. But Australian booksellers, who operate in a market rich with unique colonial-era publications, Indigenous-language texts, and small-press runs, say the destruction is being felt especially hard in their niche. The small volumes that built local literary identity are disappearing not into libraries, but into garbage bins and recycling plants.\n\n## \u2018More Than Just Objects\u2019\n\nThe phrase \u201cmore than just objects\u201d has become a rallying cry for booksellers who see a difference between digitizing a book for preservation and destroying the physical artifact itself. A rare book is not identical to its digital scan; provenance, binding, marginalia, and the physical history of each copy add layers of meaning that a database cannot capture. Sellers argue that the current wave of AI training data collection treats each volume as disposable raw material, stripping away the very qualities that made it cherished.\n\nBook dealer and preservation advocate Anne Malcolm, speaking to local outlets in Melbourne, described the practice as a \u201cquiet catastrophe.\u201d She noted that in recent months, several of her colleagues had seen collections sold off cheaply to intermediary companies that offered no public explanation for their purchases. Only later did the booksellers learn that the materials were being bulk-scanned for training datasets, with the original copies being thrown away. \u201cWe have lost complete runs of community magazines, government gazettes, and even hand-annotated novels,\u201d Malcolm said. \u201cOnce that paper is gone, it is gone forever.\u201d\n\nThe issue has drawn comparisons to the notorious book-burning campaigns of history, though without the ideological fervor; this time, the motivation is commercial and technological. But the effect can be equally permanent. Rare-book dealers note that even a mediocre copy of a scarce title is often the only remaining evidence that the work existed, and once that copy is pulped, any errors in the digital scan become irreversible. A missing page, a torn plate, or a faint pencil note cannot be restored.\n\n## The AI Training Gold Rush\n\nThe demand for AI training data has grown exponentially since the release of large-scale language models. Technology companies have scoured every corner of the internet, and when that well runs dry, they turn to physical archives. Books have long been considered a high-quality source for training data because they contain complex, long-form, and edited prose that is often more structured than web text. Scanning printed works at scale allows firms to build datasets that improve the \u201creasoning\u201d and \u201cstyle\u201d of their models.\n\nThis has produced a massive appetite for books, especially titles that are not protected by strong digital licensing or that are otherwise difficult to access online. Rare-book dealers report that intermediaries, sometimes acting on behalf of overseas AI companies, have approached them with offers that are hard to refuse financially. Yet the same dealers are growing wary of the opaque supply chains. Once a book is handed over, there is no guarantee that the original will survive the process, and no legal requirement forces the buyer to preserve it.\n\nIntellectual-property attorney Fiona Gallagher, who specializes in cultural heritage, said that the practice raises serious legal and ethical questions. \u201cThere is no clear rule that says scanning a book entitles you to destroy it,\u201d Gallagher explained. \u201cCopyright law protects the expression of ideas, but it does not ensure the physical preservation of a work. The AI companies are not acting illegally if they lawfully bought the books, but the moral residue is enormous.\u201d\n\n## Calls for Transparency and Regulation\n\nAustralian bookselling associations have now begun to organize, demanding that tech companies disclose how many physical books they are processing, what happens to the originals after scanning, and whether any rare or culturally significant materials are being singled out for destruction. Some have proposed a moratorium on bulk book destruction until an independent oversight body is created. Others want AI companies to adopt a \u201cpreservation pledge,\u201d under which any scanned work would be offered to a national library before the physical copy could be disposed of.\n\nThe Australian Library and Information Association (ALIA) has added its voice to the concern, urging public institutions to treat any unusual bulk purchase of old books as a potential red flag. A spokesperson for ALIA said that the organization was exploring whether existing heritage laws could be extended to protect private collections that are purchased for commercial data extraction. \u201cIf a book is unique in the world, its destruction is not a private matter,\u201d the spokesperson said. \u201cIt is the loss of a public good.\u201d\n\nThe idea is not unprecedented. Several national libraries have digital-priority policies that allow mass scanning for preservation, but they almost always keep the original copies. The difference is that libraries, unlike AI firms, are governed by public accountability and long-term preservation mandates. Booksellers would like to see similar ethical standards applied to any private entity that seeks to digitize physical books for any purpose, including AI.\n\n## Preserving the Unscannable\n\nA further complication is the nature of what is being lost. Many of the books targeted for AI training are not simply old; they are physically idiosyncratic. Annotations, library stamps, binding threads, glued-in corrections, and even the acidification of paper all convey information that no flatbed scanner captures. For researchers studying print culture, book history, or the history of reading, the destruction of such copies is a blow to scholarship well beyond the loss of the text itself.\n\nAustralian universities have already reported anecdotal cases of researchers losing access to rare copies that had previously been held in private collections. One historian of colonial Australia, who asked not to be named, said she had planned to examine a unique missionary dictionary printed in a South Australian town in the 1860s. The dealer who had owned it told her it had been sold to a data company. The scanner produced a broken PDF that had been assembled, ironically, back into the machine of an AI system. The original was gone.\n\n## A Global Conversation\n\nThe debate in Australia echoes growing concerns worldwide. In Europe, rare-book sellers have expressed similar alarm over the disappearance of early printed works from the antiquarian market. In the United States, librarians have questioned whether AI companies should be allowed to take advantage of mass digitization exceptions that were designed for preservation, not commercial model training. The difference in Australia, however, is the country\u2019s relatively small supply of rare materials and its massive geographic spread, which makes it difficult to know how much has already been lost.\n\nBooksellers say they are not opposed to AI training itself, nor to the digitization of collections in general. Many have embraced scanning as a way to share their catalogs with a wider audience. What they oppose is the destructive, unaccountable nature of current practices. As one Sydney dealer put it, \u201cWe let people copy our books, but we don\u2019t sell them to be incinerated. If AI wants our knowledge, it should find a way that doesn\u2019t erase the original.\u201d\n\n## What Happens Next\n\nAustralia\u2019s book trade is now pressing for a formal industry code of conduct, and the question has attracted preliminary interest from the Australian Competition and Consumer Commission (ACCC), which has broad powers to investigate unfair business practices. The ACCC has not yet confirmed whether it will open an inquiry, but booksellers are compiling a dossier of reported cases, including records of companies that acquired entire private libraries and refused to disclose the final fate of the materials.\n\nIn the meantime, some dealers are adopting self-imposed safeguards. A growing number are adding clauses to sales contracts that require buyers to preserve the physical book after scanning or to offer it to a designated archive. Others have begun quietly building \u201clast-copy registries,\u201d informal databases of unique or extremely rare titles, so that if a book disappears, the community knows what has been lost.\n\nYet those measures cannot undo the damage that has already occurred. For every book that makes a checklist, dozens more are processed without notice. The booksellers\u2019 alarm is not simply the reaction of a niche industry defending its livelihood; it is the warning of custodians who understand that knowledge lives in the object as well as the words. As Anne Malcolm put it, \u201cA book is a physical memory. You do not throw away your grandmother\u2019s letters just because you read them once.\u201d\n\nThe coming months will determine whether Australian regulations — perhaps even a national review of the AI data-supply chain — will protect those memories before the last copies meet a machine, a scan, and then a shredder.","category":"Technology","imageQuery":"rare books destroyed AI training Australia"}}