Japan’s leading anime and game studios are pushing back against OpenAI. Studio Ghibli, Bandai Namco, and others accuse the company’s video model Sora 2 of using Japanese works without consent.
In a letter from the Content Overseas Distribution Association (CODA), the group urged OpenAI to stop training the model on member content, warning that much of Sora 2’s output “closely resembles Japanese material” and that, under Japan’s law, “prior permission is required.”
The letter that put Sora on notice
Acting on behalf of Studio Ghibli, Bandai Namco, Aniplex, Square Enix, Kadokawa, and Shueisha, the association filed a formal request urging OpenAI to stop using Japanese works to train its AI model.
The letter goes further than protest. CODA warned that even the act of replication during machine learning could amount to infringement, and dismissed OpenAI’s opt-out system as no protection under Japan’s copyright law. The group also pressed the AI company to address every rights-holder claim tied to Sora 2’s outputs, showing that Japan’s creators expect answers, not algorithms.
Is Sora’s opt-out system enough?
OpenAI says it’s giving creators control. Under Sora 2’s current policy, studios and copyright holders can choose how — or if — their works can appear in the model’s generations. The system was designed to let rightsholders set boundaries, echoing the company’s rule that real people and public figures can’t be recreated without consent.
Yet in practice, Sora 2’s outputs have already featured SpongeBob, Pikachu, and other recognizable characters, showing how easily copyrighted worlds slip into its feed. Even with Disney reportedly opting out, many fan-made clips continue to circulate, like Mario breaking free from his game, anime scenes “in the style of Studio Ghibli,” and countless mashups born from prompts that never named the originals.
For Japanese publishers, the issue isn’t about adjusting settings after the fact but about being asked first. Where OpenAI frames its policy as “granular control,” Japan sees a gap: consent that arrives too late to count, and a system that still lets imitation travel faster than permission.
Reaction is not prevention
The limits of Sora’s safeguards became clear long before Japan’s protest. In the US, OpenAI halted the use of Martin Luther King Jr.’s likeness only after users created and shared deepfake videos that twisted the civil rights leader’s image. The company and the King Estate later issued a joint statement promising stronger “guardrails for historical figures.”
The episode reflects Japan’s concern that OpenAI’s consent systems activate only in reaction, not prevention. Whether it’s a public figure or a piece of art, the pattern is the same; control comes after the controversy, not before it.
While debates over copyright and consent continue, the AI race isn’t slowing. OpenAI, Anthropic, and Cohere are expanding new specialist teams aimed at bringing their models into everyday use.