The New Yorker's Francesca Mancino this week surfaced something that deserves more oxygen than it has received: Anthropic is attempting to destructively scan all the books in the world. Destructive scanning means cutting the spine, feeding pages through a sheet feeder, discarding the physical object. The goal is training data. The method is what it sounds like. This is not a metaphor.
What Gets Lost When the Object Goes
The argument for digitization has always been access: more people can read a scanned book than a shelved one. But destructive scanning is a different proposition. The physical book, with its marginalia, its binding decisions, its paper degradation, carries information that no scan captures. David Senior at SFMOMA's library has written and spoken extensively about the book as a primary object rather than a container for text, a distinction that becomes urgent when the physical container is being fed through a machine to extract only the text layer. Walter Benjamin's argument about the aura of the original, the irreducible thing that reproduction cannot carry, was about photography and painting. Anthropic is making it about the codex.
The Governance Question Nobody Is Asking Loudly Enough
A 2026 Nature commentary by Yuan Shen argued that patients must be placed at the center of medical AI governance, not as an afterthought. The same argument applies to cultural heritage: the institutions whose books are being scanned, and the publics those institutions serve, are not at the center of decisions about what gets destroyed and what gets modeled. Brewster Kahle at the Internet Archive has spent decades arguing that digitization done right is a public good, but that public accountability is the condition, not a bonus. Anthropic's approach appears to invert that: private benefit, public loss, and the conversation happening mostly in magazine culture sections rather than in regulatory hearings.