The advancement of text-to-video AI models like OpenAI’s Sora 2 has been rapid and unstoppable. However, the technology may have just hit a major roadblock from some of the most influential creators in the world. Companies behind some of the world’s most beloved Japanese IP—including Studio Ghibli, Square Enix, and Bandai Namco—have formally demanded that OpenAI immediately stop using their copyrighted content to train the Sora 2 video generation model.

CODA, which represents a powerhouse lineup of Japanese intellectual property (IP) holders, issued a formal demand to OpenAI. Their intervention follows a surge of Sora 2 output that clearly mimics, or closely resembles, famous Japanese characters and art styles—a trend previously seen when GPT-4o launched and flooded the internet with “Ghibli-style” images. The move could be the origin of a serious threat to the way generative AI gathers its data.

CODA Demands OpenAI stop using Japanese IP for Sora 2 AI text-to-video model

The core of the legal conflict lies in the difference between legal systems. CODA asserts that Sora 2’s outputs prove the model used their members’ content as machine learning data. This is where the infringement occurs, according to the group.

OpenAI has publicly discussed an opt-out system for IP holders. This means companies must actively request that their content be excluded from the training data. CODA, however, argues that this system violates the established legal precedent in Japan. Under Japanese copyright law, content creators must generally grant prior permission—an opt-in system—before their works can be used. Using an opt-out policy, CODA argues, does not excuse the liability for copyright infringement in the first place.

This is a key distinction. CODA is not just complaining about the generated videos that look like Studio Ghibli. They are actually claiming that the very act of copying the works for machine learning constitutes a copyright violation.

What happens next?

CODA’s formal request demands that OpenAI ensure its members’ content is not used for future AI training without explicit permission. It also insists that OpenAI sincerely address all copyright claims related to Sora 2’s outputs.

While there’s no immediate legal action announced, the tone from the trade group is firm. This conflict represents a major global challenge to the way large AI companies build their models. If major tech companies are forced to switch to an opt-in system, the volume and diversity of data available for training models could be severely limited. This would force a complete shift in AI development strategy.

The post Ghibli & Bandai Lead Charge Against OpenAI Over Sora 2 Training Data appeared first on Android Headlines.