The battle between Hollywood and generative artificial intelligence (AI) took an important procedural turn this month in one of the most closely watched AI copyright cases pending in the United States. While some headlines suggested that Disney and Universal were broadly ordered to disclose their own internal AI use, the actual discovery ruling is more nuanced. The court permitted only limited discovery into AI-related issues tied to market harm, technological safeguard, and the scope of requested relief, while rejecting most of Midjourney's broader requests regarding the studios' internal AI development and use.
Although procedural, the ruling highlights several of the most significant legal questions facing businesses that develop or deploy generative AI.
The Case
Disney Enterprises, Inc., et al. v. Midjourney, Inc.
United States District Court, Central District of California
Case No. 2:25-cv-05275 (JAK-AJR)
Following the initial lawsuit filed by Disney and NBCUniversal in June 2025, related claims by Warner Bros. were later consolidated into the litigation, making the case one of the largest copyright actions against a generative AI image platform.
The plaintiffs allege that Midjourney unlawfully:
- Used copyrighted works to train its image-generation models;
- Generated highly recognizable depictions of copyrighted characters such as Darth Vader, Mickey Mouse, Shrek, Yoda, Bart Simpson, and others; and
- Continued producing allegedly infringing outputs after receiving notice from the studios.
The complaint seeks damages and injunctive relief restricting Midjourney's ability to generate copyrighted characters without appropriate safeguards.
Midjourney's Defense
Midjourney has advanced several defenses expected to become recurring themes in AI copyright litigation.
- AI Training as Transformative Use
Midjourney argues that training AI models on billions of publicly available images is fundamentally different from reproducing those images.
Midjourney's position is that:
- The model learns statistical relationships rather than storing expressive copies;
- The training is transformative rather than substitutive; and
- The resulting model does not contain copies of the underlying works.
This issue mirrors legal questions already pending in numerous AI copyright lawsuits involving text, music, images, and source code.
- Output Versus Training
Perhaps the more difficult issue is distinguishing between the legality of training on copyrighted works and the legality of outputs that closely resemble protected characters.
Even if courts ultimately conclude that training itself qualifies as fair use, highly recognizable outputs may still create independent copyright liability. Such a distinction may ultimately become one of the defining principles of AI copyright law.
- Plaintiffs' Own Use of AI
Midjourney also argued that Disney and Universal themselves use generative AI internally.
Its theory is not simply "everyone does it." Rather, Midjourney contends that evidence regarding the studios' own AI practices could bear on issues such as:
- Alleged market harm;
- Available technological safeguards;
- Industry standards; and
- Equitable defenses.
The court largely rejected Midjourney's request for sweeping discovery into the studios' internal AI development and use, finding much of it irrelevant or disproportionate, but allowed targeted discovery tied to narrower issues such as market harm, safeguards, and the scope of requested relief. It is important to emphasize that the court did not broadly compel production of all Disney and Universal AI usage; instead, it substantially narrowed the requested discovery.
The "Unclean Hands" Argument
Muc commentary has described Midjourney's position as an "unclean hands" defense. That characterization deserves caution.
Unclean hands is an equitable doctrine requiring misconduct directly related to the claims at issue, not merely similar industry behavior. Instead, Midjourney appears to be using discovery to challenge:
- The plaintiffs' claimed harm;
- The feasibility of technological protections;
- The reasonableness of the requested injunction; and
- Factual assertions regarding AI development and deployment.
Whether those arguments ultimately succeed remains to be seen.
Why This Case Matters
This litigation extends well beyond Hollywood. The court's eventual rulings may affect a wide range of companies that develop, license, deploy, or rely on generative AI tools, including:
- Software companies;
- Marketing agencies;
- Health care organizations;
- Financial institutions;
- Publishers;
- Advertising platforms; and
- Enterprise AI vendors.
Several critical questions remain unresolved:
- Is training on copyrighted material itself an infringing act?
- Is AI training protected as fair use?
- When does an AI-generated output become an infringing derivative work?
- What technical safeguards must AI developers implement?
- How should courts distinguish inspiration from unlawful copying?
Practical Takeaways for Businesses
Regardless of the outcome, organizations deploying generative AI should consider:
- Maintaining written AI governance policies;
- Documenting approved AI use cases;
- Evaluating training data provenance where feasible;
- Implementing controls to reduce generation of copyrighted material;
- Reviewing vendor representations regarding intellectual property rights;
- Allocating contractual responsibility through indemnification and risk-shifting provisions;
- Maintaining privilege-aware documentation around AI legal review, vendor diligence, and approved use cases; and
- Monitoring rapidly evolving litigation and regulatory developments.
Businesses should also recognize that legal risk may arise not only from AI training but from downstream outputs generated by employees, contractors, or customers using AI.
Looking Ahead
The Midjourney litigation represents one of the first major tests of how traditional copyright principles apply to generative AI image models.
While the discovery dispute generated significant attention, it does not establish that Disney or Universal engaged in wrongful conduct, nor does it determine whether AI training constitutes copyright infringement. Those substantive questions remain for future proceedings.
What the case does underscore is that courts are beginning to scrutinize not only how AI models are trained but also how copyrighted content is generated, commercialized, and governed. The eventual decisions may help define the legal boundaries of generative AI for years to come.
Cases and Authorities to Watch
- Disney Enterprises, Inc., et al. v. Midjourney, Inc., No. 2:25-cv-05275 (C.D. Cal.)
- Consolidated Warner Bros. Entertainment action against Midjourney (consolidated into the Disney lead case)
- Magistrate Judge Joel Richlin's June 2026 discovery order addressing Midjourney's motion to compel AI-related discovery