Fair use is a US copyright doctrine allowing use of copyrighted material without permission when certain conditions are met. Courts weigh four factors: whether the use is commercial or educational, how much was copied, whether it's transformative (creating something new rather than substituting for the original), and whether it harms the market for the original.

AI companies lean on the transformative argument: training a model on a book is more like studying it than republishing it. In 2025, a federal judge agreed in a case involving Anthropic, ruling that scanning legally purchased books to build training data was transformative fair use.

But fair use is decided case by case — and Anthropic also faced a separate settlement over using pirated copies of those same books. That inconsistency captures why AI copyright law remains unsettled: the doctrine was written for human authors borrowing from each other, not machines ingesting libraries at industrial scale.