The Disney-OpenAI Agreement Transforms the Battle Over AI Copyrights

On Thursday, Disney and OpenAI unveiled an agreement that seemed improbable not too long ago. Beginning next year, OpenAI will have the right to use Disney’s iconic characters, including Mickey Mouse, Ariel, and Yoda, in its Sora video-generation model. Disney will invest $1 billion in OpenAI, granting its employees access to the company’s APIs and ChatGPT. This scenario raises questions—unless Disney is facing a challenge it can’t overcome.
Disney has a long-standing reputation for aggressively protecting its intellectual property. Together with Universal, another powerhouse in IP, it filed a lawsuit against Midjourney in June for allegedly infringing on classic film and TV characters. Just before the announcement of the OpenAI partnership, Disney reportedly issued a cease-and-desist letter to Google, claiming copyright violations on a “massive scale.”
At first glance, Disney’s collaboration with OpenAI seems contradictory, especially while confronting its competitors. However, it appears that Hollywood is taking a page from the media publishing industry’s playbook regarding AI, opting for licensing agreements where possible and resorting to litigation when necessary. (WIRED, for instance, is owned by CondĂ© Nast, which formed a partnership with OpenAI in August 2024.)
“I think that AI companies and copyright holders are starting to realize that neither side will achieve a complete victory,” says Matthew Sag, a law and artificial intelligence professor at Emory University. While many of these legal matters are still pending in courts, it seems that model inputs—the training data used for these models—are typically protected under fair use. However, this agreement pertains to outputs—what the model generates based on the user’s prompt—where IP holders like Disney hold a considerably stronger position.
Reaching an agreement on outputs helps clarify a range of complex, potentially unsolvable issues. Even if a company instructs an AI model not to create, say, Elsa at a Wendy’s drive-through, the AI might still conjure up something similar, as it retains knowledge of the character. A user could also devise prompts that effectively generate Elsa without directly naming her. Legal scholars refer to this dilemma as the “Snoopy problem,” but in this context, it could just as easily be termed the Disney problem.
“Confronted with this increasingly apparent reality, it is reasonable for consumer-facing AI firms and entertainment titans like Disney to consider licensing deals,” notes Sag.
