AI Companies Destroying Millions of Books for Training Data

• AI companies are purchasing millions of physical books in bulk and destroying them after scanning for training data.
• This practice has led to a surge in demand for obscure and out-of-print books.
• A federal judge ruled that destructive scanning of legally purchased books can be considered fair use.
• Anthropic will pay authors approximately $3,000 per book in a $1.5 billion copyright settlement.
• Concerns exist about the permanent loss of rare and uncommon books due to this practice.

AI companies are engaging in a practice akin to book burning, anonymously purchasing millions of physical books in bulk, destroying them after scanning their content for AI training data. This surge in demand has significantly boosted sales for booksellers, particularly for obscure and out-of-print titles, leading to fears that rare books are being permanently lost.

A federal judge recently ruled that the destructive scanning of legally purchased books can qualify as transformative fair use, a decision that has implications for ongoing copyright litigation involving major AI firms like OpenAI and Meta. However, in a separate development, Anthropic has agreed to a $1.5 billion copyright settlement, paying thousands of authors approximately $3,000 per book for using pirated copies to train its AI model, Claude.

The practice has drawn criticism, with some viewing it as a race to preserve human-authored knowledge before it is overshadowed by AI-generated text. Prominent figures like Elon Musk have publicly opposed the destructive scanning method, advocating for more traditional scanning techniques to preserve rare books.