AI Companies Are Buying Antique Books, Ingesting Their Contents to Train Models, and Then Destroying Them at Incredible Scale, Even If Almost No Copies Remain

Artificial intelligence companies are increasingly sourcing physical books to train their models, a practice that leads to the systematic destruction of printed material. Because these developers require data untouched by modern internet content, they seek out older volumes that remain free from AI-generated text. This demand has turned antique books into industrial feedstock rather than preserved knowledge.

Anthropic remains a prominent example of this trend, having previously reached a massive settlement after shredding physical copies to scan the contents for model training. The process often uses high-speed cutting machines to separate pages for industrial imaging equipment. These firms rely on the first-sale doctrine to avoid copyright hurdles, framing the conversion of physical to digital as a transformative act.

Database providers now facilitate bulk orders that range from one thousand to one million books at a time. Smaller booksellers report sudden spikes in demand for random inventory, often suspecting that anonymous tech labs are the buyers. These distributors frequently promise secrecy to their clients to manage the negative perception surrounding the practice.

Booksellers acknowledge a moral conflict regarding these sales. While the influx of cash clears out old inventory, the loss of rare or out-of-print editions weighs on those in the trade. Reports indicate that this behavior is spreading globally, with rare book dealers in the Netherlands also seeing signs of bulk purchasing by technology firms looking to feed their algorithms. The pursuit of clean data continues to override the preservation of physical literature.