Note: The full text of this discovery order was not publicly available at time of publication. This summary is based on Law360 reporting and federal court docket records.
Background
Disney Enterprises, Universal Pictures, and Warner Bros. Entertainment sued Midjourney, Inc. — the company behind the popular AI image-generation service Midjourney — for copyright infringement, alleging that Midjourney used billions of copyrighted images to train its generative AI models without authorization. The case is one of several pending AI copyright lawsuits in the Central District of California brought by major entertainment studios and artists against generative AI companies.
Midjourney’s defense strategy includes the argument that the plaintiff studios themselves use and develop AI systems that may have been trained on copyrighted content — in other words, the studios are doing the same thing they are suing Midjourney for doing. To pursue this defense, Midjourney filed a motion to compel the studios to produce discovery related to their own AI development and use, including source code for one of the studios’ AI models (referenced in the docket as “v8 training source code”). The studios resisted production, arguing the requested materials were either irrelevant or protected from disclosure under the work product doctrine.
The Court’s Holding
Magistrate Judge A. Joel Richlin partially granted and partially denied Midjourney’s motion to compel. The court found that some of the studios’ AI-use data was appropriate for production — the studios’ own internal AI practices bear at least some relevance to the issues in the case, and they cannot categorically shield all information about their AI systems from discovery. At the same time, Judge Richlin denied Midjourney’s broader discovery requests, finding that portions of the requested materials were either irrelevant to the claims at issue or fell within the protection of the work product doctrine (a legal shield that protects documents and communications prepared in anticipation of litigation).
The upshot: Midjourney obtained partial discovery into the studios’ AI use — including at least some data related to the studios’ own AI model training — but the court drew a line protecting materials the studios developed as part of litigation strategy or analysis.
Key Takeaways
- AI copyright defendants can discover plaintiffs’ own AI use: This ruling signals that defendant AI companies in copyright lawsuits can obtain at least some discovery about how plaintiff studios and creators use AI themselves. The “you do it too” defense, while not a complete answer to copyright infringement, may generate relevant evidence about licensing practices, industry standards, or equitable considerations.
- Work product doctrine limits — but does not eliminate — AI discovery: Courts will still require plaintiffs to produce some AI-use data, but materials prepared specifically in anticipation of litigation can be withheld. AI companies seeking broad discovery of competitors’ systems should expect scope limitations.
- Studios face two-front AI war: The major studios are simultaneously suing AI companies for infringement while developing their own AI tools, creating litigation exposure on both fronts. Discovery orders like this one can force them to disclose information about their own AI development that they would prefer to keep confidential.
Why It Matters
The Disney v. Midjourney case is one of the most closely watched AI copyright disputes in the country. The central question — whether training a generative AI on copyrighted images constitutes copyright infringement — has not yet been definitively answered by any appellate court. But the procedural developments in the case are already shaping how AI copyright cases are litigated: defendants can pursue discovery into plaintiffs’ own AI use, the work product doctrine has real limits in this context, and courts are willing to require studios to open their own AI operations to some scrutiny.
This discovery order may matter beyond this specific case. As AI copyright litigation proliferates, both plaintiffs and defendants will look to this ruling for guidance on how much of a party’s own AI use is discoverable in copyright infringement proceedings. If studios must consistently disclose their own AI model training data in litigation, it creates significant litigation risk for any major IP holder that also develops AI tools — a category that increasingly includes most major entertainment companies.