Analysis
Midjourney is asking a federal judge to force Disney, Universal and Warner Bros. Discovery to disclose the details of their own internal AI use, escalating its defense strategy in the copyright lawsuit the three studios filed against the AI image generator starting in 2025. The studios allege Midjourney's tool lets paying subscribers generate unauthorized images of famous copyrighted characters -- Darth Vader, Shrek, Batman, Superman and Homer Simpson among them -- because Midjourney trained on the studios' intellectual property without permission.
Midjourney's core legal defense is that training AI models on copyrighted images falls under fair use, an argument still being tested across multiple ongoing AI copyright cases industry-wide. But its more aggressive recent move is procedural: pushing to compel the studios to reveal their own internal AI research, training datasets, model weights, and board-level presentations about how they use AI in filmmaking, storyboarding and marketing. This is a classic "unclean hands" defense -- arguing the plaintiffs are engaged in comparable conduct to what they're suing over.
A magistrate judge already ruled on part of this dispute in June, limiting Midjourney's discovery rights to only the studios' consumer-facing AI applications rather than their full internal AI operations. Midjourney's lawyers filed a motion this week asking Judge John Kronstadt to overturn that limitation, arguing the studios should have to disclose far more about how AI is used behind the scenes in their production and marketing pipelines.
“This is a classic "unclean hands" defense -- arguing the plaintiffs are engaged in comparable conduct to what they're suing over.”
This fight sits within a much larger wave of AI copyright litigation -- The New York Times v. OpenAI, Getty Images v. Stability AI, and various music-label suits against Suno and Udio all raise variations of the same fair-use question. What makes the Midjourney case distinctive is the discovery-reciprocity angle: rather than purely defending its own training practices, Midjourney is trying to make the litigation costly and reputationally uncomfortable for the studios themselves by exposing their internal AI use.
The numbers in context: this is one of the highest-profile AI copyright suits given the recognizability of the characters involved, and a ruling either way could influence settlement dynamics across the dozens of other pending AI-training copyright cases, many of which involve smaller plaintiffs with less leverage than three major studios.
For AI founders building generative tools trained on any copyrighted corpus, this case is worth tracking closely regardless of outcome -- both the fair-use ruling and the discovery precedent it sets will shape how aggressively future plaintiffs (and defendants) can demand disclosure of AI training and usage practices from each other.
The bear case: even if Midjourney wins the discovery fight and exposes uncomfortable studio AI practices, that outcome doesn't resolve the underlying fair-use question, and Midjourney could still lose on the merits of unauthorized character generation regardless of what the studios' own AI use looks like.
What to watch: Judge Kronstadt's ruling on the discovery scope, any depositions or document production that follows, and whether other AI-copyright defendants adopt similar reciprocal-discovery strategies against plaintiffs with their own undisclosed AI operations.
