What’s Happening
Warner Bros Discovery has sued Midjourney, a popular AI image generator, claiming it “brazenly stole” copyrighted characters to train its AI. We’re talking Batman, Superman, Wonder Woman, Bugs Bunny, Scooby-Doo—the whole Warner Bros cartoon character lineup.
The allegation? Midjourney trained its AI on these copyrighted characters without permission and then let subscribers generate high-quality images of the characters “in every imaginable scene.”
Why It Gets Interesting
There’s been a lot of attention on the mining of respected sources of information to train generative AI, prompting lawsuits by the New York Times, Encyclopedia Brittanica, and others. AI heavyweight Anthropic recently settled with an association of publishers and authors for $1.5 billion because it downloaded a half-million pirated books to feed its AI model.
We’re now seeing more attention to art and other creative works that are protected by copyright. Right-of-publicity claims by celebrities – such as we saw in the video gaming world – are sure to come as well.
Warner Bros claims Midjourney’s unauthorized use of copyrighted characters was intentional: The company allegedly blocked users from generating videos of infringing images, then quietly lifted that protection last month. Disney and Universal sued Midjourney in June over characters like Darth Vader, Bart Simpson, and Shrek.
Midjourney’s position? Training AI on copyrighted works is “transformative fair use” that helps ensure “the free flow of ideas and information.” This is going to be a legal battle over fundamental questions about AI, creativity, copyright, and free speech in the digital age.
The Mediator’s Take
From a dispute resolution perspective, here’s what’s fascinating:
The Stakes: Midjourney reportedly hit $300 million in revenue in 2024 with nearly 21 million users. That’s a lot at risk.
The Precedent: How courts rule on AI training could reshape entire industries. With each new method of copying or using works created by others has come resistance, litigation, and, eventually, new ground rules – we saw this with the development of photocopy machines, audio cassette recorders, home VCRs, and online platforms for sharing video and music.
The Pattern: Authors, news outlets, record labels—everyone’s suing AI companies right now. This isn’t getting resolved quickly or quietly, so mediation and arbitration are great alternatives to the courtroom.
What’s Next?
Warner Bros wants damages, profits, and an injunction. But the bigger question: Can you borrow something to teach a machine, or is that fair use? Does it make a difference is your source for the copyrighted works is an infringer, as in the Anthropic case? Are you “in the clear” if you train AI using copyrighted works that have been made accessible by the copyright owner? Do platforms such as YouTube have liability if AI trains without restriction on the countless hours of copyrighted video uploaded by movie and TV studios, record companies, and everyone with a cellphone?
What’s your take—is AI training on copyrighted works theft or innovation?
