AI companies accused of buying and destroying books for training data

Investigations have revealed that artificial intelligence companies are reportedly purchasing massive quantities of physical books to train large language models, often resulting in the destruction of the original copies.

Documents from a lawsuit against Anthropic suggest the company engaged in “Project Panama,” involving the purchase of millions of books, removing covers, and scanning pages before destroying the originals. Similarly, an investigation by 404 Media used an Apple AirTag to track a shipment of 1,000 books from a seller on Biblio. The tracking led to an Amazon warehouse in Las Vegas (codenamed VGT3), where workers reported receiving large shipments of books specifically to strip ISBN barcodes and covers before permanent destruction.

While Amazon stated they purchase books through commercial channels to “help the development and improvement of our products and services,” they declined to comment on the destruction of materials. Authors, historians, and bibliophiles have criticized these practices, noting that the mass acquisition of rare and antique books for data scraping threatens the preservation of unique historical records.