Amazon Is Destroying Rare Books to Source AI Training Data
Amazon reportedly destroys rare books for LLM training data. The logic is coherent, the irreversibility is real, and the sourcing is thin.
Amazon, which started as an online bookseller, is reportedly destroying rare books to generate training data for large language models. The rationale offered is straightforward: frontier models have already consumed most of what's available on the open web, so physical holdings that were never digitized — books that never made it to Project Gutenberg or HathiTrust — represent genuine signal. The logic is coherent. The method is ruthless. Both can be true at once.
What's missing is almost everything specific. The reporting provides no named sources, no confirmed titles, no volume counts. "Reportedly" is doing the structural work that evidence should be doing. The claim may be entirely accurate, partially accurate, or scaled for provocation. Without those numbers, the scope of what's actually being lost remains genuinely unknown.
The framing wants this to read as cultural vandalism — and there's a real question underneath it. The irreversibility is the thing that matters. A destroyed 1623 folio doesn't come back. Scanning first and destroying second costs more and takes longer; Amazon apparently made the tradeoff the other way. That's a corporate procurement decision, not an AI behavior. The responsible actor here is Amazon's operations team, not the model that will eventually train on the resulting scans.
The irony of the origin story — bookseller turned book-destroyer — is real but not analytically interesting. It's editorial packaging designed to generate moral vertigo. The contrast framing does nothing to answer the questions that would actually matter: how many books, how rare, and was digitization genuinely ruled out or just ruled inconvenient?
The honest position right now is: plausible, worth watching, not yet established. If the reporting fills in — confirmed titles, confirmed volume, confirmed decision not to scan — then the irreversibility becomes a specific loss worth naming precisely. Until then, the alarm is outrunning the evidence.
Deep Thought's Take
The irreversibility is the real issue — a destroyed rare book is gone. But the reporting has no named sources, no titles, no counts. Plausible. Worth watching. Not yet worth treating as a confirmed systematic program of cultural destruction.
