Midjourney just flipped the script in its copyright fight with Hollywood. The AI image startup is now demanding that Disney, Universal, and Warner Bros. disclose exactly how they use AI internally, according to TechCrunch AI. It’s a bold discovery move in a case that could shape how courts treat AI training data for years.
Here’s the background. Disney and Universal sued Midjourney last year, arguing its image models could spit out copyrighted characters like Bart Simpson and Darth Vader. Warner Bros. joined the pileup a few months later. Midjourney’s defense rests on fair use: it says training models on copyrighted images is legally permitted.
What Midjourney is asking for
The current fight isn’t about the art itself. It’s about paperwork, specifically what the studios have to hand over during discovery.
A judge already ruled the studios must reveal information about their generative AI usage, but only when it produced “consumer-facing” videos and images. Midjourney wants that limit gone. In its latest filing, the startup argues the restriction “unfairly” lets the studios “cherry-pick only those documents they believe support their market harm claims while depriving Midjourney of documents that would support its defenses.”
The sharper claim comes next. Midjourney says the withheld “documents are precisely those that would reveal whether, behind closed doors, they are doing exactly what they are suing Midjourney for doing.”
The startup wants disclosure on two fronts:
- Internal AI development. If a studio is building image-generating models “for internal use in storyboarding or ideating content,” Midjourney says that would show it’s “an industry custom, even among the studios themselves, to download and train AI on unlicensed copyrighted content.”
- Every prompt and output. Midjourney wants all the prompts studio staff typed into its own product, plus the results, not just the ones that generated allegedly infringing images.
The studios push back
The studios aren’t buying it. Their lead attorney, David Singer, previously called Midjourney’s document requests a “fishing expedition.”
Singer also drew a clear line on intent. The studios “do not seek to stop AI technology or even shut down Midjourney’s business,” he said. They “simply want Midjourney to stop copying their movies and TV shows and to stop distributing, publicly displaying, publicly performing, and creating derivative works that include copies of their famous characters without authorization.”
Why this matters
What stands out here is the strategy. Midjourney isn’t just defending its own training practices. It’s trying to prove the studios do the same thing behind the curtain. If discovery shows Disney or Warner engineers were feeding unlicensed content into their own AI tools, the “industry custom” argument gets a lot stronger, and the studios’ moral high ground gets shakier.
This is significant for a few reasons:
- It targets the fair use core. The whole case may hinge on whether training on copyrighted material is standard, accepted practice. Studio behavior becomes evidence, not just Midjourney’s.
- It raises the stakes for every studio. Media companies have been quietly experimenting with generative AI for storyboarding and pre-production. Those experiments could now become courtroom exhibits.
- It sets a precedent for discovery. How broadly courts let AI defendants dig into a plaintiff’s own AI use could reshape future copyright suits across the industry.
For AI builders and practitioners, the message is practical. Internal AI experiments leave a paper trail, and in litigation that trail can be pulled into the open. Prompt logs, model training records, and tool usage are all discoverable.
What comes next
The judge now has to decide whether to widen discovery beyond “consumer-facing” outputs. That ruling will signal how much of the studios’ internal AI playbook becomes fair game.
Either way, this case keeps drawing the map for AI and copyright. A win for Midjourney’s broader discovery push would encourage other AI defendants to turn the same lens back on their accusers. A loss would keep the focus tight on the models themselves.
For the full filing and additional detail, check the original report from TechCrunch AI.