Reports indicate that some AI companies are anonymously and massively purchasing physical books through intermediaries, disassembling their spines, scanning them into training datasets, and discarding the original copies to extract high-quality text written by humans prior to the proliferation of generative AI. This demand has already impacted the secondhand book market, with one bookseller noting weekly sales rising from around 20 to hundreds of copies, though expressing concern that rare and out-of-print books may be permanently destroyed after digitization. Similar practices have emerged in Anthropic’s Project Panama, which previously employed destructive scanning methods to digitize millions of books.

Last summer, a federal judge in San Francisco ruled in Bartz v. Anthropic PBC that scanning legally acquired physical books into digital copies—even after destroying the originals—constitutes transformative fair use; similar rulings have since appeared in cases involving OpenAI and Meta. However, the same district court this week approved a $1.5 billion copyright settlement for Anthropic, requiring the company to compensate thousands of authors approximately $3,000 per book for training Claude on pirated works. Musk stated he has instructed his team to preserve rare books and adopt non-destructive scanning methods.