Two U.S. district court decisions on AI training copyright in June 2025 addressed distinct uses on the records before them, leaving the question dependent on the facts in later cases.
The rulings came in separate cases: Bartz v. Anthropic and Kadrey v. Meta. In Bartz, a federal judge analyzed the use of books to train models apart from Anthropic's storage of pirated copies in a central library.
In Kadrey, the court said it had to decide the case based on evidence presented by the parties. Neither decision set a nationwide rule for every AI training practice.
A May 2025 report from the U.S. Copyright Office (USCO) said the result can vary by use, with some AI training uses expected to qualify for fair use and others not. The office listed unlawful acquisition of training material, a model's functionality and deployment, and effects on markets for copyrighted works as relevant considerations. The report discusses those factors.
