This resurfaced this week (some are calling it “AI book burning”), and I think the legal angle is more interesting than the outrage framing, so here's a neutral breakdown.
What's documented: To build a training corpus, Anthropic bought millions of physical print books and “destructively scanned” them — cutting off the bindings, scanning the pages, and discarding the physical copies. It even hired someone who'd previously worked on Google's book-scanning program to acquire books at scale.
The counterintuitive part: they destroyed the books partly for legal reasons. Buying a physical copy and digitizing it — without keeping a duplicate — looks much more like legal “format shifting” than downloading pirated files. In Bartz v. Anthropic, Judge William Alsup ruled that training on legally purchased, destructively scanned books was fair use, while using pirated books was not. Anthropic later agreed to pay ~$1.5B to settle the piracy claims.
So the odd takeaway is that shredding books you bought became the *legally cautious* option.
The open question people are debating: it's arguably fine for bulk used paperbacks (the text survives in countless copies), but what about rare or out-of-print editions where each physical copy actually matters? Once those are cut up, the object is gone even if the words live on as data.
I wrote up the full breakdown with the case details here: https://thebotpost.com/ai-news/ai-firms-destroying-millions-books-train-models
Is destructive scanning of purchased books a reasonable price for training data, or should rare/irreplaceable editions be off-limits?
[link] [comments]