Court Rules AI Training Is Fair Use Except for Pirated Book Copies
A federal judge issues a split ruling in Bartz v. Anthropic, declaring that most AI training methods are highly transformative fair uses, while explicitly ruling that using pirated books crosses the line into copyright infringement.
A federal judge issues a landmark split decision in the case of Bartz v. Anthropic, addressing whether using copyrighted books to train large language models constitutes fair use or infringement. Judge William Alsup rules that Anthropic’s use of the books to train its Claude AI models is "spectacularly" transformative, marking one of the first substantive judicial decisions on this highly debated legal issue.
The court divides Anthropic's actions into distinct categories, finding two of them to be fair use. The judge determines that using purchased books to train AI for new text outputs is a highly transformative fair use, and rules that digitizing legally purchased print books to replace discarded physical copies also qualifies as fair use.
However, the court draws a strict line regarding the origins of the training data by ruling that Anthropic's use of pirated book copies is direct copyright infringement. While this summary judgment resolves the fair use question for these specific practices, numerous other issues from the lawsuit remain unresolved and are set to proceed to trial.