Irish bookshop and online retailer Kenny's told The Irish Times this week that they have been receiving unusual book orders via email in the middle of the night, requesting upwards of 5,000 often obscure titles. The pattern in Galway fits a global story of AI companies buying physical books to feed the models that underpin chatbots, in a practice the industry calls "destructive scanning".
The process sees the spines sliced off thousands of paper books, the pages digitised, and then the whole lot destroyed and pulped. The image unsettled many readers - but it is just a glimpse into larger shifts in the economics of literature and publishing, including a recent US lawsuit against AI firm Anthropic, which alleged the company violated authors' rights by using their work without permission to train the models behind Claude.
The "silent burning" of literary culture
Something about destroying books feels instinctively wrong. We think of books as holders of knowledge and artistic beauty, objects of pleasure and instruction that transmit humanity's accumulated insights through the ages. They are things to be passed along, archived, and rediscovered.
Reading Derek Scally's piece, I imagined faceless robots consuming and discarding our literary history, an image that reveals the wider pattern of destruction being meted out on the very structure of writing. The scan-and-destroy method is a kind of silent burning, erasing the fire of a culture and strip-mining the creative class.
Companies wanted this kept quiet. In an internal memo exposed in court filings, an executive at Anthropic wrote that "Project Canada is our effort to destructively scan all the books in the world - we have not long been doing it. We don't want it to be known that we are working on this."
The legal fight over training data
The Anthropic lawsuit, brought by a group of authors, accused the company of violating ownership rights by using books without permission. The judgment pointed out that the company could have negotiated with publishers and paid a fair price - after all, "users and corporate clients pay to use Claude, generatingnt over $1 billion in annual revenue." Instead, it chose deliberately to "avoid the legal/practice/business slog" of paying for people's rights.
That case centered on books taken from pirate websites. No compensation was awarded for work that had been scanned and destroyed. The judgment stated that Claude had "created no exact copy, nor any substantial knock-off," but also noted the company "could have purchased books, but it preferred to steal them."
The wider picture is that AI-generated books are flooding the market and harming working writers. Researchers found authorities that the number of ebooks published monthly on Amazon has tripled since ChatGPT launched, while quality measured by star ratings worsened. Earlier this month, journalist Kashmir Hill found an AI-generated autobiography of herself for sale on Amazon at a novelty counter and reported it. The 90-page book contained all the hallmarks of AI slop, including no consciousness and a hallucinated meeting with the wrong stranger.
The economics of writing are shifting
There are four key inputs needed to build AI: chips, data centres, energy, and data - and data means the output of human creative labour. AI companies are deciding they are willing to pay for the first three but not the fourth. That free labour is the books, articles, and fiction that flood the training sets. But if you destroy the human layer, what is left to train the machine on?
Sales of actual books are dropping. Both fiction and non-fiction sales have fallen in every category except children's literature, according to Publishers Weekly. In the UK, a Cambridge University study found that half of published novelists say AI is likely to replace their work entirely, and 85 per cent expect their future income to be substantially reduced.
Why this matters for writers
This is not a distant policy debate that will resolve itself. Book sales are falling while algorithmically generated competitors are multiplying at a quality of stroy. The legal system has yet to provide a consistent standard protecting writers from having their work scanned and exported into AI models without compensation, and so far, courts have sided mostly with AI firms.
Writers now face a market where they are expected to produce work that competes against a flood of cheap AI slop. Understanding how AI works and its limitations can help writers demonstrate where authored work retains value. Those who learn to work AI tools as editors rather than replacements - while documenting their original human creative labour - may have a stronger case for retaining the income stream that remains.
Your membership also unlocks:
