OpenAI defends model pretraining as fair use

Where this statement fits

OpenAI says pretraining on authors' books is fair use; authors and publishers urge the court to reject it

OpenAI's September 4, 2026 summary-judgment motion argues that pretraining its models on authors' books was a transformative fair use that caused no legally cognizable market harm. The same day, the Authors Guild and the other class plaintiffs moved for partial summary judgment, asking the court to rule that OpenAI's training copies, its downloads of books from pirate sites and the transfers of books between OpenAI and Microsoft were not fair use. The authors argue that OpenAI copied their books to build models that generate competing works and deprived them of revenue from an existing market for training licenses. OpenAI argues that training extracts unprotectable statistical patterns for new uses and that the plaintiffs' dilution and licensing theories of harm are not legally cognizable. On October 29, the Association of American Publishers and three other publishing and licensing groups asked the court for leave to file a brief supporting the plaintiffs.