What Cox v. Sony Means for the AI Copyright Cases — Starting With Midjourney
July 30, 2026
Four months after the Supreme Court reversed a $1 billion jury verdict and narrowed the rules for secondary copyright liability, we're getting our first real test of how that decision plays out in the AI cases that have been working their way through the courts. Midjourney has moved for partial judgment on the pleadings in the copyright suit brought against it by Disney, Universal, and Warner Bros., asking the court to dismiss the contributory infringement theory within the case, arguing the studios haven't pled what the Court's decision in Cox Communications, Inc. v. Sony Music Entertainment now requires.
The motion doesn't touch the studios' direct infringement claim, which Midjourney continues to contest separately on fair use grounds, and it doesn't touch the vicarious liability theory either. What it targets is whether the studios have adequately pled that Midjourney intended its service to be used for infringement under Cox’s narrower test.
What Cox Actually Decided
The Cox case arose from a different but potentially analogous context. A record label accused an internet service provider of doing too little to stop its subscribers from pirating music. Courts had long recognized contributory copyright liability where a defendant knew of a user's infringement and either induced it, offered a service designed for infringement, or materially contributed to it by having the means to stop it and failing to do so. The Supreme Court's unanimous decision effectively eliminated that last path.
The Court’s limits on contributory liability could tip the scales for Midjourney. Consider that Cox had received more than 163,000 infringement notices over a two year period and terminated only a small fraction of the accounts involved. A jury found it liable for contributory infringement and awarded $1 billion in damages, and the Fourth Circuit affirmed. The Supreme Court reversed unanimously but while seven justices joined Justice Thomas's opinion adopting a categorical two-part test, Justices Sotomayor and Jackson disagreed with the part of the majority’s opinion that seemingly foreclosed common law theories such as aiding and abetting in cases of indirect infringement.
Contributory copyright liability has long required that a defendant intended its service to be used for infringement, a principle that traces back to the Court's 2005 decision in MGM Studios v. Grokster. What's new is how narrowly the Court now says that intent can be shown. Under Cox, there are only two paths—inducement, meaning the defendant affirmatively encouraged infringing use, or tailoring, meaning the service is incapable of substantial or commercially significant noninfringing uses. What Cox eliminated is a third path several circuits had allowed for years—that knowledge of infringement, combined with a failure to take reasonable steps to stop it, was enough on its own. The Court held that contributory liability cannot rest merely on a provider's knowledge of infringement paired with inadequate steps to prevent it.
Why This Was Always Going to Land on the AI Cases
Cox was decided in a broadband context, but the doctrinal question it resolved—what counts as intent when a platform's users, not the platform itself, generate the infringing content—sits at the center of many of the pending AI copyright cases. Those cases have spent the past two years focused overwhelmingly on the training inputs question—what was copied to build the model, and whether that copying was fair use. Midjourney's motion is a sign of where the fight might be moving next, toward a second and largely separate question—when a user's own prompt produces an output that resembles a copyrighted work, on what basis, if any, is the platform responsible for that output.
The studios' complaints against Midjourney allege, in addition to their direct infringement claim over training, that Midjourney knew its service was being used to generate outputs resembling their characters and failed to take available steps to prevent it. That just might come too close to the secondary liability theory the Supreme Court just eliminated, and the correspondence between counsel in the lead-up to this motion suggests both parties recognize the shift. After Midjourney argued in its April letter that Cox foreclosed what commentators have called the “knowledge-plus” theory, the studios responded that they were no longer pursuing contributory infringement on a knowledge-and-inaction theory. Where the parties still disagree is whether the studios' inducement theory has any merit and whether the contributory infringement claim should move forward.
The Explore Page, and What It Does and Doesn't Show
The studios’ inducement theory rests largely on Midjourney's “Explore” page, a gallery of user-generated images with an accompanying search function, which the studios' complaints describe as intended to advertise Midjourney's ability to reproduce their characters. The complaints illustrate this with screenshots showing manual searches for terms like “Storm Trooper” and “Superman” returning generated images resembling those characters.
Midjourney's motion pushes back on the legal significance of that evidence rather than the images themselves. Inducement, under the Cox standard, reaffirming MGM Studios v.Grokster, requires an affirmative message encouraging infringement, not merely a feature capable of being used that way, Midjourney argues the Explore page's default gallery and keyword search apply generally across all user content, no different from a search bar on any social platform, rather than being directed at reproducing any particular copyrighted work. It also argues, more pointedly, that the studios haven't alleged a single instance of one of their characters appearing in the gallery in the ordinary course, as opposed to appearing as a result of searches the studios ran themselves.
Whether a curated or algorithmically organized feed crosses the line from neutral technology into something a court will treat as affirmative encouragement is a question Cox itself never had to answer, since the ISP in that case wasn't accused of curating or promoting anything.
The Transactional Read
For companies building or licensing AI products, the emerging lesson here isn't that platforms are now insulated from liability. It's that generative AI raises a genuinely new question under copyright law—who is responsible for a given output, the platform or the user who prompted it, and how a product surfaces, ranks, or displays user-generated outputs raises different infringement questions than the earlier training-data disputes. Cox has raised the bar for what a plaintiff must show to hold an AI company liable for infringing outputs prompted by its users. A service that is purely functional, one a user directs without the platform steering the outcome, sits in a materially stronger position after Cox than one where a court finds the platform's own curation choices did the steering.
That distinction, between a service that is user-prompted and content-neutral versus one that curates or promotes specific outputs, is worth building into product design, terms of use, and content-moderation decisions well before any dispute arises. It's also a useful lens for reviewing indemnification and content-licensing provisions in agreements involving any generative AI tool with a public gallery, feed, or discovery feature.
A hearing on Midjourney’s motion is set for September 28. It's no longer just about the training data—it's increasingly about who bears the risk for the outputs. We'll know more about where the court draws that line then.