Studio Ghibli And Bandai Lead Legal Attacks On OpenAI Regarding Sora 2 Training Data

JAKARTA - The wave of resistance to the training of the generative AI model is now coming from Japan. Three creative giants "Studio Ghibli, Square Enix, and Bandai Namco" officially demanded OpenAI stop using their works in training the Sora 2 generative video model.

Through the CODA copyright industry association (Content Overseas Distribution Association), the three of them accused OpenAI's opt-out' system of violating Japanese copyright laws requiring permits before use (opt-in).

In recent months, OpenAI's text-to-video model Sora 2 has amazed the world by its ability to create realistic animations from simple texts. However, a number of results are thought to mimic Studio Ghibli's signature style and iconic characters from the Square Enix game to Bandai Namco. CODA considers this resemblance as not a coincidence, but evidence that its members' work is used in training data.

A similar phenomenon has occurred before when the GPT-4o model was busy producing Ghibli-style pictures, which raised concerns about ethical and legal boundaries in the use of copyrighted works to train AI.

This dispute is rooted in differences in legal approaches between the United States and Japan. OpenAI has been implementing an opt-out policy, where copyright holders must actively request that their work be excluded from AI training.

CODA strongly rejects the policy, insisting that Japanese law requires an explicit opt-in system from the owner of the work before it is used. That is, if OpenAI has copied and processed work without permission to train Sora 2, the action itself is considered a copyright infringement.

The use of the optim-out system is not a justification for copyright infringement, "said the official statement of CODA. They also demanded that OpenAI guarantee that its members' work will not be used in the next AI training without written approval.

Although CODA has not announced official legal steps, its stance marks a major confrontation between Japan's creative industry and the global tech giant. If this pressure succeeds in forcing companies like OpenAI to switch to an opt-in system, then the AI development landscape could change drastically.

Training data sources will shrink drastically, and future AI models may lose the diversity of visual styles that have been the main attraction. But on the other hand, these changes could strengthen the position of artists and content creators, ensuring they are no longer victims of data mining without permission.

This case is not just a matter of Ghibli or Japanese games. It has the potential to be the first global fight to actually test copyright law limits in the generative AI era between human creativity and artificial intelligence learning from the work itself.