US court rules on AI training and pirated books in Bartz v. Anthropic
A US federal judge has held that training an AI model on lawfully bought books was fair use, but that keeping pirated copies was not.
On 23 June 2025 Judge William Alsup of the US District Court for the Northern District of California issued an order on fair use in Bartz v. Anthropic, a case brought by the authors Andrea Bartz, Charles Graeber and Kirk Wallace Johnson against the AI company Anthropic.
The court granted summary judgment to Anthropic on two points. It held that using books to train large language models was fair use, describing the use as highly transformative. It also held that scanning print books the company had bought, to create digital copies for its internal library, was fair use.
The court reached a different conclusion on the millions of books Anthropic had downloaded from pirate sites and kept in a central library. It refused summary judgment on those copies, finding that this was not fair use, and ordered that the issue, and any damages, go to trial.
Why it matters: this was among the first US rulings to apply fair use to AI training on books. It is a first-instance decision under US law and does not bind UK courts, but it draws a clear distinction between how material is used and how it is obtained. For authors and publishers, the source of any copies used for training has become a central question.
Sources: Wiggin and Dana: Bartz v. Anthropic, first court decision on fair use in LLM training; The Authors Guild: Mixed decision in Anthropic AI case.