Midjourney has asked a federal court to compel Warner Bros. Discovery, Disney, and Universal Studios to reveal how they train their own AI models. The image generator argues the studios are using the same copyrighted-data practices they're suing to stop, a move that could undermine their infringement claims.

Last year the studios sued Midjourney for generating protected characters like Superman and Batman. Midjourney counters that scraping public images constitutes fair use and has requested the studios' AI business plans, training datasets, model weights, and even board-meeting presentations. A magistrate judge ruled in June that the studios need only disclose consumer-facing AI applications, shielding most internal documents.

Midjourney attorney Bobby Ghajar told the court that evidence of the studios training on copyrighted works goes to the heart of the fair-use and unclean-hands defenses. The federal judge's ruling on this motion will shape what internal AI development records become admissible in future copyright litigation across the industry.