OpenAI shares legal arguments that AI training is fair use

On Oct. 30, 2026, OpenAI published excerpts from court briefs arguing that training AI on copyrighted work can be legal. The post is not a court ruling. It collects arguments from creators, academics, accessibility advocates and legal experts.

What the courts have said

Fair use is a legal doctrine that can allow use of copyrighted material when the use is transformative and does not substitute for the original. Two district court decisions in June 2025 leaned that way. In Bartz v. Anthropic (June 23), training was fair use on the record, but pirated books kept in a central library were not protected. In Kadrey v. Meta (June 25), the plaintiffs had not shown market harm.

Those rulings are not a nationwide rule. The U.S. Copyright Office said in a May 9, 2025 pre-publication report that training can be transformative, but how a model works, how it is deployed, unlawful acquisition and market effects all matter. That guidance is not binding law.

What is still open

Questions remain about how training data is obtained, how it is stored and what happens to the market for original work. Ohio State University Libraries reviewed the rulings and the Anthropic settlement in a March 20, 2026 update, which points to continuing uncertainty. No next court date was supplied. For anyone who uses AI tools or publishes original work, more decisions are likely and could change the picture.