Court Rules Scanning Books for AI Training Is 'Fair Use'

AI labs are adopting a strategy of buying physical books published before 2022 in bulk and scanning them, in order to prevent Model Collapse caused by 'AI Slop' generated since 2022.

Key Process and Background:

• Securing Data Quality: To secure high-quality data that is 100% written and edited by humans, rather than synthetic text found on the internet.
• Scanning Method: Industrial hydraulic cutters are used to remove the binding of books, after which high-speed scanners rapidly digitize thousands of pages, and the paper waste is shredded.
• Legal Ruling: In 2025, a federal court (Judge William Alsup) ruled that 'Format-shifting'—converting legally purchased print media into training data—constitutes Transformative Fair Use.

However, the court explicitly stated that downloading and using illegal pirated libraries such as Books3 cannot be protected under fair use. Meanwhile, concerns are also growing in the book market about the rapid disappearance of rare books due to demand for AI training.