Japanese CODA orders OpenAI to cease all usage of material to Sora 2 AI training

When it comes to generative AI, the services available have long faced criticism for not truly being "generative" but rather derivative, since the models rely on existing data for training, using it as a kind of inspiration when creating new assets, code, or solutions.

That's why training data is absolutely crucial for companies like OpenAI, Meta, and Google. But other content creators are starting to catch on. The Japanese industry association Content Overseas Distribution Association (CODA), which represents many of the country's largest content producers, including Bandai Namco, Square Enix, Studio Ghibli, and others, has now issued a formal notice to OpenAI, asking the company to stop using any of their collective works as training data for Sora 2, which launched on October 1.

Through Sora 2, users quickly began creating videos featuring copyrighted IP's such as Mario, One Piece, Dragon Ball, Demon Slayer, Cloud from Final Fantasy, and many others.

CODA isn't just demanding that OpenAI stop using their content for training and output generation, they also argue that Sora 2 has already been trained so extensively on copyrighted material that the content it produces itself constitutes copyright infringement.

"CODA considers that the act of replication during the machine learning process may constitute copyright infringement," they say.

OpenAI previously stated that certain studios were given a week's notice to opt out of Sora 2 training, but the company has not disclosed who was contacted.