Midjourney demands studios reveal their own AI training to prove fair use

The Midjourney case sits at the intersection of two conflicting legal principles: copyright protection for creative works and the emerging doctrine of fair use in AI training. The studios' complaint centers on Midjourney's training method—feeding the AI system publicly available images, including copyrighted ones—to enable users to generate images of protected characters. Midjourney's counterargument hinges on what it calls "industry custom and practice": the claim that the major studios themselves engage in the same conduct (scraping and unlicensed copying of millions of images for AI training). If true, this symmetry could undermine the studios' position, since copyright's fair use doctrine weighs factors like the market effect and industry norms. A mid-June ruling already blocked Midjourney's initial discovery request, but the company's motion to review signals it believes access to the studios' training practices is essential to mount a fair use defense.

The broader significance lies in the unresolved status of fair use in generative AI. Prior court decisions have sometimes ruled that AI training is transformative and thus fair use, but those rulings have been inconsistent and often turned on specifics like proof of market harm or the rights holder's failure to establish why fair use should not apply. With hundreds of similar cases pending worldwide, each ruling will incrementally shape copyright law and the rights of artists, rights holders, and AI companies. The outcome could affect not only Midjourney but the entire business model of training large AI systems on internet-scale data.