AI labs are turning old books into training data and the paper disappears

Futurism, citing 404 Media, describes AI companies and intermediaries buying physical books in bulk, scanning their contents for model training and in some cases destroying the books. The article also points to Anthropic, which court materials say used a cutting machine to remove book spines and industrial scanners to digitize pages. Anthropic separately reached a $1.5 billion settlement with authors over claims involving books and training data.

The cleanest data is on shelves, not the web

Pre-LLM books are valuable to model companies because they are not contaminated by generated text from the newer internet. ISBNdb, as quoted in the article, sells that advantage directly: physical books published before 2022 are framed as clean of modern poisoning tools and AI slop.

According to 404 Media, ISBNdb helps arrange bulk purchases from 1,000 to one million books per order. It also offers buyer anonymity, which says plenty about how sensitive the practice has become.

For models, a physical book is data with better provenance

Strategically, this reverses the old logic of digitization. Libraries used to be digitized for human access. Now physical collections are being converted into input material for models because the web is losing data purity. Older, rare and foreign-language books suddenly gain value as training material.

For publishers and booksellers, the trade is uncomfortable. One seller in the article described jumping from no more than 20 books a week to hundreds, while worrying that unusual or out-of-print titles were disappearing. The money arrives immediately. The cultural loss is counted later.

A legal copy does not settle the inventory problem

The legal argument relies on first-sale doctrine and fair use, as the article describes for physical books. But legality and stewardship are not the same thing. If anonymous bulk orders lead to scarce books being cut apart, society loses specific copies, not just abstract content.

The next fight is transparency and the line on destruction

The signal to watch is whether rules emerge for rare and unavailable books, or whether the market keeps running quietly through intermediaries. AI companies like calling data fuel. With old books, the fuel is also the archive.

Lilith's verdict

When a model eats a book and leaves only cutter dust behind, the issue is bigger than copyright. It is a library scene where someone sold the last chair because the wood had a good price.