Midjourney’s copyright fight with Hollywood has moved into a new phase: the AI image company is trying to force Disney, Universal, and Warner Bros. Discovery to disclose more about how they use artificial intelligence inside their own businesses.
The dispute is procedural, but it matters. A June discovery order limited Midjourney mostly to information about the studios’ consumer-facing AI applications. Midjourney is now asking U.S. District Judge John A. Kronstadt in the Central District of California to widen that scope, according to recent reporting from Variety and summaries of the filing by Engadget and VP Land.
Midjourney wants materials that could include AI business plans, research reports, training datasets, model weights, production-use documents, and board presentations about AI strategy. The studios, which accuse Midjourney of enabling mass infringement of protected characters and scenes, have resisted that broader look into internal AI work.
A copyright case becomes an AI governance fight
The underlying case began in June 2025, when Disney, NBCUniversal, and DreamWorks sued Midjourney over AI-generated images that allegedly reproduced characters and other protected material from franchises including Star Wars, Marvel, Pixar, and DreamWorks films. Warner Bros. Discovery later brought related claims that were consolidated into the same litigation.
The studios’ complaint focuses heavily on outputs: images and video generated from prompts that allegedly display recognizable copyrighted characters, not just the training process behind the model. A Georgetown Law analysis of the original filing noted that the studios sought injunctive relief requiring copyright protections against reproduction of their intellectual property, alongside damages and other remedies.
Midjourney’s defense has leaned on a familiar AI-industry argument: training on publicly available material can be transformative and therefore fair use. The discovery fight adds a sharper question. If major studios also train or operate generative AI systems using copyrighted works, Midjourney argues, that evidence could bear on fair use, market harm, available safeguards, and equitable defenses.
That does not mean the studios have been found to have done anything improper. It also does not decide whether Midjourney infringed copyrights. The point is narrower: how much of a plaintiff’s own AI practice can become evidence when the plaintiff is suing an AI company over training, outputs, safeguards, and market harm?
The consumer-facing line may be hard to hold
The June order, as summarized by Shumaker’s legal analysis, was more limited than some headlines suggested. The court allowed targeted discovery related to market harm, safeguards, and requested relief, while rejecting much of Midjourney’s broader request for internal AI-development information.
That leaves a practical divide between public-facing products and behind-the-scenes AI systems. Consumer-facing applications are easier for courts to connect to market harm or public substitution. Internal tools used in production, visual development, storyboarding, marketing, or workflow automation are harder to frame, especially if they never ship directly to viewers.
Midjourney wants to collapse that distinction. Its lawyers argue that internal training practices and AI governance choices could show what the entertainment industry itself treats as normal, necessary, or technically feasible. If a studio uses copyrighted material to evaluate or train internal models, for example, that could become relevant to the industry-standard and fair-use arguments Midjourney wants to make.
The studios’ narrower framing is straightforward: the case is about Midjourney allegedly copying, displaying, distributing, performing, or creating derivative works from studio content. On that view, internal studio AI use is not a license for a third-party platform to generate recognizable characters for subscribers.
Why other companies should watch the ruling
The next discovery decision could affect more than entertainment companies. Many businesses are now both AI users and potential copyright claimants. A publisher may object to model training while using AI for newsroom tools. A software vendor may sue over code use while deploying coding assistants internally. A brand may demand image-generation guardrails while using generative AI for ad concepts, mockups, or customer-facing personalization.
That dual role makes AI documentation more sensitive. Policies, training-data notes, vendor diligence, internal model evaluations, board decks, and product-roadmap material may not stay purely internal if a company later becomes involved in AI litigation. The Midjourney fight is a reminder that AI governance is not just about compliance or security; it can also become evidence.
For companies adopting generative AI, the practical lesson is not to avoid AI tools entirely. It is to document approved uses, separate experimental work from production workflows, understand vendor claims about training data and indemnity, and preserve privilege-aware review around sensitive legal analysis. Companies should also be careful about making sweeping public claims against AI training if their own internal practices depend on similar assumptions.
The case still has a long road ahead. The court has not resolved the central questions of whether Midjourney’s training practices are protected fair use, whether particular outputs are infringing derivative works, or what safeguards an AI image or video platform must implement. But the discovery fight shows how quickly AI copyright cases can expand from model behavior into corporate AI practice.
If Judge Kronstadt broadens discovery, Hollywood may have to reveal more about how generative AI is being used behind studio walls. If he keeps the current limits, the case will remain more tightly focused on Midjourney’s own system, its outputs, and the public AI products the studios make available. Either way, one of the biggest AI copyright fights in the U.S. is now testing how much a company’s internal AI use matters when it asks a court to police someone else’s.
