OpenAI and Disney Just Ended the ‘War’ Between AI and Hollywood with $1 Billion Sora Deal

“AI companies will either need to actively filter user prompts and model output to ensure they don’t accidentally show Darth Vader, or reach agreements with copyright holders to gain permission to produce videos and images of Darth Vader,” Sager told wealth. “The licensing strategy is more of a win-win.”

The three-year agreement gives OpenAI the rights to bring hundreds of Disney-owned characters to Sora and ChatGPT Image. Disney will also receive equity warrants and become a major customer of OpenAI while deploying ChatGPT internally.

The deal itself will be a “revenue share,” Sager said.

“OpenAI hasn’t figured out the revenue model yet,” Sager said. “So I think, in a way, having this just an investment deal simplifies it. For Disney … (OpenAI) is going to find a way to be profitable at some point, and (Disney will) take a piece of that pie.”

Why this deal matters: The “Snoopy problem”

For more than a year, the biggest legal threat to large-scale generative AI has focused on what Sager calls the “Snoopy problem”: training powerful generative models without some level of memory is extremely difficult, and copyrighted characters are particularly vulnerable because copyright protects them in the abstract.

Sager is careful to outline a key distinction. AI companies are not granting training rights on copyrighted works; they are licensing creative rights output Otherwise it will constitute infringement.

That’s because AI companies have “very good reasons” to train their models on unlicensed content, Sager said. Two recent court decisions involve Anthropic selection and Yuan strengthens these arguments.

The real stumbling block, Sager said, has been outputnot training. If a model accidentally produces a frame that looks too much like Darth Vader, Homer Simpson, Snoopy, or Elsa, the fair use defense starts to suffer.

“If you do remember too much, if that memory ends up in the output, then your legitimate use cases start to break down,” Sager said.

While it’s impossible to get enough text to train an LLM (“it would be a billion-dollar” deal, Sager says), if you have the right partner, you can build an image or video model entirely from licensed data. That’s why deals like Disney’s are crucial: they turn previously illegal output into legal output, regardless of whether the training process itself qualifies as fair use.

“The limiting principle is essentially about whether these models replicate most of the work from the training data in day-to-day operations,” Sager said.

Sager said the deal is also a hedge against Hollywood lawsuits. For Disney, the news is ‘very bad’ for Midway Are suing for copyright infringement, Because it insists on treating OpenAI’s licensing agreement as a “responsible” benchmark for artificial intelligence companies.

This is also a signal about the future of AI data

In addition to copyright risks, the deal exposes another trend: the depletion of high-quality, unlicensed data on the public internet.

in a blog postSugg wrote:

“The low-hanging fruit of the public Internet has already been picked,” he wrote. “To get better, companies like OpenAI will need access to data that others don’t have. Google have Youtube; OpenAI now owns the Magic Kingdom. “

This is the core of what he calls the “data scarcity argument.” The next leap in OpenAI model quality may require exclusive content partnerships, not more scraping.

“By entangling itself with the world’s largest intellectual property holder, OpenAI has made itself an integral part of an industry that threatens to sue it,” Sager wrote.

AI and Hollywood spent three years caught in a cold war Regarding training data, portrait rights and copyright infringement. With Disney’s $1 billion investment, that era appears to be coming to an end.

“This is the template for the future,” Sager wrote. “We are moving away from an all-out war between AI and content and towards a negotiated carving up of the world.”