Anthropic purchased and destroyed millions of physical books to scan their contents for training its Claude AI chatbot [1].
This practice highlights the aggressive pursuit of raw data by AI developers and the potential loss of physical cultural heritage to avoid legal costs.
Internal documents released in a 2024 lawsuit show the company engaged in "destructively scanning" during 2023-2024 [1, 3]. The process involved purchasing millions of books [1], removing their bindings, and scanning the pages before discarding the physical copies [1, 2, 3]. This method allowed the company to obtain vast amounts of text while avoiding the cost and complexity of negotiating copyright licenses [2, 4].
The scanning took place at facilities operated by contractors in the U.S. [2]. According to reports, the collection included rare and antique titles [2]. Some of the books were described as "less common" titles [3].
"Anthropic bought millions of physical books to train its AI chatbot Claude," CBC News said [1].
The Guardian editorial staff said the company apparently found the destruction of books easier than dealing with copyright in its quest for training data [4]. While some reports suggest multiple AI companies may be involved in similar practices [2], internal documents specifically link these actions to Anthropic [1, 3].
The scale of the operation is significant. While some commentary suggests an attempt to scan nearly every book in existence [4], verified reports confirm the number of books purchased and destroyed reached into the millions [1].
“Anthropic purchased and destroyed millions of physical books to scan their contents for training its Claude AI chatbot.”
The destruction of physical texts to feed large-language models represents a shift in how AI companies acquire data. By converting physical assets into digital tokens and destroying the originals, firms can bypass traditional copyright frameworks that apply to digital distribution. This creates a tension between the technical requirements of AI scaling and the preservation of physical archives, particularly regarding rare or antique works that cannot be replaced once destroyed.
