Booksellers Warn AI Companies May Be Destroying Rare Volumes for Training Data

Booksellers and preservationists allege that certain AI companies are purchasing rare and fragile books solely to destroy them in order to create cheap digital copies for model training. The practice reportedly involves feeding spines into wood chippers and scanning torn pages before discarding the remnants, a method favored for its speed and low cost. Google holds a 2009 patent for a non-destructive scanning technique that could protect such volumes, but the slower, more labor-intensive process is less appealing to firms eager to ingest massive amounts of high-quality text.

Studies have shown that even the patented method can miss pages or distort text due to page curvature, and operational glitches have been observed. The Internet Archive, which works with libraries to digitise aging collections, stresses that careful handling is essential for rare books, underscoring that the current destructive trend jeopardises cultural heritage.