Disney/OpenAI Deal Rewards Act First, Justify Later

Disney Executive Chairman Bob Iger (Photo by Charley Gallay/Getty Images for Disney)

Getty Images for Disney

OpenAI and Disney announced that they entered a three-year deal which allows Sora 2 users to draw from more than 200 Disney, Marvel, Star Wars and Pixar characters to generate short, user-prompted videos that they can share with others – subject to certain guardrails. Under the deal, OpenAI’s ChatGPT will also allow users to generate images involving Disney’s IP.

One lesson is clear from the Disney and OpenAI deal: OpenAI’s strategy of act first, justify later is working.

By building a model based on third-party IP, and daring rightsholders to object, OpenAI has created a self-reinforcing cycle: drive engagement through infringement, create value for their new service, use this value to fund settlements, and license with companies after the fact. Although in this case, OpenAI gets a $1 billion cherry on top.

Sora 2’s Flip-Flop

I recently wrote about OpenAI’s launch of its generative AI video platform, Sora 2, which was launched with an opt-out model, meaning that intellectual property owners were invited to “opt out” if they did not want Sora 2 to use their IP. Social media platforms were immediately bathed with a massive wave of user-generated videos featuring popular, valuable IP such as SpongeBob SquarePants, Darth Vader, South Park characters and Elsa from Frozen.

This, in turn, sparked a stern reaction from Hollywood and action by various authors, IP owners and authors’ guilds to opt-out of Sora 2 and demand a justification for Sora’s behavior. To be clear, offering an AI model trained on existing IP and that generates derivative works featuring that existing IP on an opt-out basis is by no means legal, so OpenAI was taking a big risk by trying it.

The response to the backlash was quick. Within three days of the launch, Sora 2 changed its model from opt-out to opt-in, along with a not entirely thorough attempt to cleanse the model of output featuring IP from owners who had not opted in, including those owned by Disney.

An Investment and a Two-Way License

One would expect the grant of such a broad license from Disney to command OpenAI to pay a large license fee to Disney.

Here, interestingly, the payment is the other way around. Under the deal, Disney will invest $1 billion in OpenAI to build new products, tools and experiences, and Disney employees will be provided with their own ChatGPT.

By this move, it would appear that Disney believes it will be able to recoup and profit from its $1 billion investment by using OpenAI’s technology in future productions, operations, and businesses. In return, OpenAI has secured the right to use a stable of the most valuable IP for its output.

OpenAI Gets $1 Billion, Google Gets a C&D Letter

As an underscore to its new deal, the day before the announcement, Disney sent a cease and desist letter to Google, asserting that Google was infringing Disney’s IP “on a massive scale” by using its copyrighted works to train and generate output on its AI tools, including Veo, Imagen and Nano Banana.

Some say Google’s Gemini 3 has surpassed OpenAI on a technological level. If true, is Disney’s $1 billion investment in OpenAI a wise investment? Has Disney set aside another $1 billion for Google?

Days after receiving Disney’s letter, Google responded by removing Disney content from its systems. Disney has sent similar letters to Meta Platforms and Character.AI and has filed a closely-watched copyright infringement lawsuit against Midjourney.

The Law Remains Unclear

Dozens of lawsuits have been filed by other IP owners – including book authors, news publications and music owners – against a myriad of AI platforms suddenly competing for space in the marketplace. As I previously reported, only two court decisions have emerged in the U.S. on the issue of whether the ingestion of copyrighted books for training purposes constitutes fair use, which is a defense to copyright infringement.

Both decisions, at the trial court level, concluded that the use of copyrighted books for AI training purposes was fair use. However, in both cases, it was found that the platforms in question, Meta’s Claude platform and Anthropic’s Llama platform, have guardrails that prevent reproductions of the copyrighted books from being generated by users. Sora 2 and Midjourney, by contrast, do not have such guardrails and in fact are designed to allow such outputs. That distinction might lead a court to conclude that the use of the IP for training in such models is not fair use, but a decision is still pending.

Disney’s deal with OpenAI not only preempts a legal decision on that question but also helps Disney’s position against Google and its other targets, since now Disney can point to an existing licensing market for its IP, which bolsters its position that the use of its IP for AI training is not a fair use since it negatively affects an existing licensing market.

IP is Fragmented

Last month, a court in Munich, Germany, held that ChatGPT’s use of song lyrics for AI training purposes and output constituted an infringement of the songwriters’ copyrights, and did not fall within the fair dealing laws or exemptions for data mining under European law. This ruling, if affirmed on appeal, could plausibly lead to a requirement that ChatGPT secure licenses for all copyrighted works ingested by ChatGPT.

Another development of note is the recent deal between actors Michael Cain and Mathew McConaghy to license their voices to AI audio company ElevenLabs for AI generated content. Notably, the OpenAI/Disney deal does not include talent likenesses or voices which Disney does not control.

More Questions than Answers

The disputes between IP owners and AI platforms thus continue to be resolved in a variety of way and on a worldwide basis. However, few, if any, ground rules have emerged, and some deals and court decisions raise even more questions.

While companies like OpenAI and Disney have, at least for the moment, come to an understanding, there is much dust to be settled, and many questions remain, including:

Are there distinctions to be made between using IP for pure training purposes as opposed to generating output that replicates the IP?

For AI models that can be used to generate replicating output, can their owners afford to license all the IP they need? Or will these deals permit only deep pockets like OpenAI and Google to compete?

Does the Disney/OpenAI deal provide a precedent where the AI system can use IP in exchange for a license to the IP owner to use the system? How will this affect competition among AI systems?

The Disney/OpenAI deal and the ElevenLabs deal highlights the fact that IPs are often a combination of images, music, voices and likenesses that may be controlled by different parties. Disney’s predominance in animation makes it a bit easier to separate these different rights while still being able to offer valuable IP.

Other studios with more live-action prevalent libraries may not be able to license their IP so easily without running afoul of the rights of actors and writers behind their assets. Might the talent guilds fill a need here?

Mash-ups, fan films, memes and other user-generated content has existed on the internet for years without the help of AI. Does the fact that AI makes fan use easier, more voluminous, or higher quality, make a difference to an IP owner?

Stay tuned.