In a court filing that reframes a sprawling copyright battle, Midjourney is asking a judge to force Hollywood’s biggest studios to open their own books on artificial intelligence. The company wants the studios to disclose how much AI they use in production — including whether any of them used Midjourney’s technology to train their own models.
The move is a counterpunch. The studios have accused Midjourney of training its image-generation systems on copyrighted film and television, a claim the company denies. Midjourney’s lawyers now argue that the accusation carries a burden of symmetry: if the studios are demanding disclosure of what went into Midjourney’s models, the studios should answer the same questions about their own pipelines. The filing asks the court to compel production of internal records on AI use, training data and vendor relationships.
The request is more than a procedural maneuver. Hollywood is simultaneously suing AI companies and adopting their tools. Studios have hired AI vendors for visual effects, dubbing and script analysis; some have experimented with generative models in development. A disclosure order would put the industry’s own AI consumption on the record for the first time, in sworn documents that could be used in other litigation and in negotiations with labor unions, which have already made AI use a central contract issue.
The underlying dispute is one of the highest-stakes tests of whether AI companies can build on the accumulated culture of the last century without paying for it. The studios argue that Midjourney’s models, trained on vast image corpora, absorbed the look and grammar of film without license. Midjourney argues that its systems transform what they learn, in the way that artists have always borrowed from what came before. The courts have not settled which view prevails, and the discovery fight is the first place where either side’s evidence will meet the rules.
Lawyers not involved in the case said the discovery fight is where this litigation will be won or lost. Copyright cases against AI companies have proliferated across the industry, and the courts have not yet produced a definitive ruling on whether training on copyrighted material is infringement. In that vacuum, the evidence gathered in discovery — what went into the models, who knew, what was licensed — becomes the substance of the case. Midjourney’s filing tries to widen the lens to include the plaintiffs’ own conduct.
The stakes extend beyond the two sides. A ruling that forces studios to disclose their AI usage would establish a template for similar disputes in music, publishing and software. Conversely, a ruling that keeps such records shielded would let plaintiffs pursue infringement claims while keeping their own practices private. The symmetry argument is attractive to courts, several attorneys said, because it treats the parties equally and avoids the appearance of one-sided discovery.
The template question matters beyond this case. Media companies have filed a wave of suits against AI developers, and the discovery battles in each will define how much of the underlying conduct becomes public. A narrow ruling here would leave each case to discover its own facts; a broad one could make AI-use disclosure a standard feature of copyright litigation. Either way, the studios’ own records — contracts with AI vendors, internal use policies, training logs — are likely to surface somewhere, because the plaintiffs’ own AI practices have become relevant in nearly every one of these disputes.
The case also lands at a delicate moment for the industry’s business model. Studios are under pressure to cut costs, and AI tools offer the clearest path — which is why unions have fought for disclosure and consent provisions in every major contract since 2023. If the studios’ own records show heavy AI use in the very areas where they accuse Midjourney of infringing, the moral high ground of the litigation gets harder to hold.
There is also a definitional problem lurking in the request. “Using AI” can mean anything from a fully generative pipeline to a spell-checker. Midjourney’s filing anticipates the dodge, asking for records that distinguish between AI-assisted work and AI-generated content, and for the names of vendors and models used. The studios will argue the request is overbroad and would sweep in trade secrets. The fight over that line is itself the beginning of the answer to the question the whole litigation poses: where the boundary of AI’s intellectual-property debt actually sits.
For Midjourney, the filing is also a public-relations move aimed at its own community of artists and designers, many of whom have mixed feelings about the company. By casting itself as the party demanding transparency, the company positions itself on the side of disclosure — a posture that plays well in a debate where opacity is the accusation most often leveled at AI firms. The judge’s decision on the request, expected in the coming months, will show whether the tactic works.
