Midjourney just turned the AI copyright fight on its head. The image startup is now demanding that Disney, Universal and Warner Bros. reveal how they quietly use AI behind closed doors. What the studios are hiding could decide who wins.
By Blumefield | July 5, 2026
A Startup Turns the Discovery Tables
For a year, the story of the Midjourney lawsuits has been told one way. Hollywood, the wronged party, versus a scrappy image generator that let subscribers conjure Darth Vader and Bart Simpson on demand. This week that story flipped. In a filing lodged on July 4, Midjourney asked a federal judge to force Disney, Universal and Warner Bros. to hand over the details of their own artificial intelligence programs, recasting the AI copyright fight as a question the studios may not want answered.
The move is aggressive, and it is deliberate. Midjourney is no longer content to defend its training data. It wants to interrogate the accusers, arguing that if the studios are quietly building or using generative models trained on unlicensed material, then the entire premise of their AI copyright claim collapses into hypocrisy. It is the legal equivalent of turning a spotlight back on the person holding it.
The case, Disney Enterprises v. Midjourney, sits before Judge John Kronstadt in the Central District of California, and the discovery skirmish now unfolding will shape far more than one company's fate. It touches every studio, every model builder and every creative worker wondering who actually owns the machinery of the next decade of filmmaking.
The Filing That Rattled the Studios
The immediate trigger is a dispute over documents. Back in June, Magistrate Judge Joel Richlin handed the studios a narrow win, ruling that they only had to produce information about generative AI when it fed into consumer-facing images and video. Internal experiments, board discussions and research prototypes stayed off limits. Midjourney's July filing seeks to overturn that limit, and the language is pointed.
The company argues the ruling 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." More bluntly, it claims the withheld files are "precisely those that would reveal whether, behind closed doors, they are doing exactly what they are suing Midjourney for doing." Midjourney attorney Bobby Ghajar wrote that if the plaintiffs "are doing the very thing they seek to punish, that evidence goes to the heart of Midjourney's fair use and unclean hands defenses."
Midjourney is not asking for a modest paper trail. It wants the studios' AI business plans, research reports, training datasets, model weights and the internal presentations executives gave their own boards about storyboarding and ideation. It also wants every prompt the studios themselves typed into Midjourney, along with the outputs, not just the handful that produced allegedly infringing images. The studios' lead lawyer, David Singer, has dismissed the request as a "fishing expedition" and insists his clients simply want the copying to stop. He has been careful to add that the studios "do not seek to stop AI technology or even shut down Midjourney's business." That distinction may prove hard to hold.
Why Unclean Hands Could Matter
The doctrine at the center of this AI copyright dispute is old and slightly obscure. Unclean hands is an equitable defense that says a plaintiff cannot obtain relief for conduct it is guilty of itself. If you sue someone for trespassing while standing on their lawn, a court may decline to help you. Midjourney is betting that the same logic applies to training data. If Disney or Universal downloaded copyrighted works to train an internal model for concept art, the argument goes, they can hardly claim moral outrage when a startup does the same at scale.
Legally, the defense is a long shot. Courts apply unclean hands narrowly, and the misconduct usually has to relate directly to the claim at issue. A studio experimenting with AI for pre-production is not obviously the same as a public tool generating thousands of branded characters for paying users. But Midjourney does not need to win the defense outright to benefit. It needs the discovery. Every internal email, every slide deck, every awkward admission that a studio trained on scraped data becomes leverage, both in court and in the far larger battle for public opinion.
That is the quiet genius of the strategy. Fair use turns heavily on market harm and on whether a use is transformative. If the studios are themselves treating unlicensed training as an industry norm, Midjourney can argue that such copying is customary rather than exceptional, weakening the claim that its own conduct is uniquely damaging. The AI copyright question stops being about one bad actor and becomes about an entire industry's habits.
Hollywood's Quiet AI Problem
The uncomfortable truth is that the studios have not been AI abstainers. Visual effects houses have used machine learning for years to age actors, clean plates and generate crowds. Concept artists lean on generative tools for mood boards. Marketing teams test poster variations at volume. The industry that publicly frets about AI theft has privately folded the technology into nearly every stage of production, which is exactly the contradiction Midjourney wants a jury to see.
This is why the discovery fight is so sensitive. A public accounting of how Disney, Universal and Warner Bros. actually use generative models could embarrass executives who have spent two years positioning themselves as guardians of human creativity. It could also complicate their relationships with the guilds and unions that extracted hard-won AI protections in recent labor deals. If the studios were quietly training on unlicensed data while negotiating those protections, the reputational cost would dwarf any single verdict.
There is a strategic irony here too. The studios sued to draw a bright line around their intellectual property. In doing so they may have invited scrutiny of their own most guarded secrets. Discovery is a two-way street, and Midjourney has decided to drive down it at speed. For readers tracking how this reshapes the creative economy, Blumefield will continue to follow the filings as they land.
What a Ruling Could Set Loose
Judge Kronstadt now faces a decision with consequences well beyond this docket. If he sides with Midjourney and widens discovery, he effectively tells every AI defendant that suing over training data opens your own practices to inspection. That prospect could cool the appetite of rights holders who assumed litigation was a one-sided weapon. It would make the next AI copyright suit a far riskier undertaking for any plaintiff with its own models to hide.
If he upholds the magistrate's limit, the studios keep their internal work sealed and preserve a cleaner narrative of victim versus infringer. But even a loss for Midjourney sharpens the central tension of this era. Courts are being asked to decide not just whether training on copyrighted work is lawful, but whether the companies loudest in their objections are quietly doing the same thing. That is a question the entire content industry would rather not litigate in public.
For now, the filing sits with the court, and the studios must respond. What began as a straightforward infringement case has become a referendum on whether Hollywood can hold AI to a standard it may not meet itself. However Judge Kronstadt rules, the AI copyright wars have entered a phase where accusers can expect to be searched as thoroughly as the accused, and the outcome will ripple through every negotiation between the people who make culture and the machines now learning to imitate it.
Primary filings and docket entries in Disney Enterprises v. Midjourney are available through the public court record and Midjourney's latest motion. Guidance on the boundaries of fair use is published by the U.S. Copyright Office. More technology and markets coverage is at Blumefield.