In recent months, a growing number of second-hand bookstores across the UK have reported bulk purchases of their inventory by companies linked to major technology firms, sparking concerns over the practice known as "destructive scanning." This process involves buying physical books, removing their pages, scanning the text for digital use, and subsequently destroying the original volumes.

One affected bookseller, Lucy Hingston from Compass Books in Dartmouth, Devon, described the situation as a “moral dilemma,” expressing distress over the destruction of what she considers valuable and sometimes rare books. Reports indicate that tech companies are targeting niche second-hand bookshops, focusing on those with online catalogues, to acquire books at low cost, sometimes consisting of obscure or specialized volumes not commonly digitized.

Documents released earlier this year revealed that Anthropic, the artificial intelligence company behind the chatbot Claude, has been engaged in destructive scanning since early 2024 under a project codenamed "Project Panama." According to internal memos, the firm spent millions purchasing millions of physical books, which were then cut from their bindings, scanned, and discarded. Previous reliance on pirated digital content for training their AI models had led to a high-profile legal challenge. The company faced a $1.5 billion out-of-court settlement with authors over copyright infringement, involving unauthorized use of millions of digitized books obtained from illegal online sources.

Under U.S. law’s "first sale doctrine," established by the Supreme Court in 1908, purchasers of physical media legally may resell, destroy, or repurpose those physical copies, a loophole the companies have used to justify their scanning practices. This allows them to train AI models on the scanned text without seeking direct permission or compensation from authors or publishers.

The practice has provoked backlash from authors, publishers, and booksellers alike. Dan Conway, CEO of the Publishers Association, condemned the approach as part of a broader pattern of copyright infringement harming creative industries and those who depend on them. Critics argue that the destruction of physical books, particularly collectible or out-of-print copies, represents a loss to cultural heritage and raises ethical questions about valuing old media solely as data sources for AI.

The companies involved defend their practices by comparing them to dissemination rather than censorship, noting the goal is to broaden access to the information contained within these printed works. Nonetheless, the physical destruction of books has drawn strong emotional reactions, with some comparing the act to historical book burnings, albeit with differing intent.

In addition to Anthropic, whistleblower accounts have revealed that large-scale scanning operations are also occurring in Amazon warehouses in the United States, where workers reportedly process thousands of second-hand volumes daily for AI training purposes.

Technological alternatives capable of scanning books without damaging their bindings exist and are reportedly employed in some cases, particularly for rare volumes. However, the destructive scanning method remains preferred by some firms as it is faster and reduces scanning artifacts like shadows.

The unfolding situation highlights ongoing tensions between emerging AI technologies’ data requirements and existing intellectual property frameworks, raising questions about sustainable and ethical approaches to training AI systems while preserving access to physical cultural materials.