Ghibli, Bandai challenge OpenAI over Sora 2 data

IP owners raise alarm

A coalition of Japanese companies represented by the Content Overseas Distribution Association, including Studio Ghibli and Bandai Namco, submitted a written request on October 27, 2025, asking OpenAI to stop using their copyrighted works to train Sora 2 without permission.

They argue the model’s output mirrors their characters and styles, raising questions about copyright infringement. The dispute shines a light on how generative AI models interact with creative content and rights-holders.

What Sora 2 does and its reach

Sora 2 is OpenAI’s flagship video and audio generation model, launched in late 2025, that allows users to generate high-fidelity media, including animations and stylised scenes.

Since its late September release, Sora 2 has generated a large volume of videos that mimic Japanese animation aesthetics and occasionally reproduce imagery evocative of well-known Japanese franchises associated with CODA members.

Critics point out that many outputs closely resemble existing copyrighted works, which is not just a case of fandom but potentially infringement. The model’s viral presence amplified the dispute.

Why Japanese IP holders object

CODA notes that many Sora 2 outputs “closely resemble Japanese content or images,” linking this to unauthorised use of copyrighted training data.

CODA says that under Japanese practice, copyright owners should give permission before their works are reused for commercial training and that an opt-out approach does not satisfy that standard in Japan.

From the studios’ perspective, using their creative content for training and then recreating it without their consent may constitute infringement. This legal and cultural friction is at the heart of the demand.

OpenAI’s opt-out policy scrutinised

OpenAI initially relied on an opt-out approach for Sora, letting rights holders request exclusion rather than requiring prior opt-in, but in early October, the company said it would introduce more granular controls in response to backlash.

CODA argues this conflicts with Japanese copyright law, which tolerates only prior permission and does not recognise retroactive “remove me” mechanisms as adequate.

The dispute centres on whether an opt-out system is sufficient when training data includes copyrighted content. If precedent shifts, training datasets may need to be constructed via opt-in rather than opt-out.

Creators demand cessation of training use

In a letter dated October 27, 2025, CODA requested that OpenAI “cease using its members’ content for machine learning without their permission” and respond sincerely to complaints.

The group emphasises that the act of training on copyrighted material and allowing recreation of identifiable styles or content might be unlawful, and so far, OpenAI has not confirmed a full halt of this usage.

The request underscores that rights-holders seek active control, and not merely downstream filtering or revenue sharing.

Examples of contested outputs

Among the viral Sora 2 outputs are scenes incorporating characters and styles from Studio Ghibli films, Bandai Namco games, and other major Japanese IPs, despite no licensing deals being announced.

One widely circulated video featured an AI-generated Sam Altman amidst Pokémon characters saying, “I hope Nintendo doesn’t sue us,” illustrating how the model enabled surreptitious use of protected characters.

These examples fuel concerns by rights-holders that the model uses their IP indirectly.

Cultural and legal impact in Japan

Japan’s parliament and legal figures are now weighing in, as the country is being urged to lead global regulation of AI-training-data practices, given its creative export profile.

The interconnectedness of Japanese media culture and global AI flows places the dispute in a national context, not just a corporate one.

If Japanese law views training data rights strictly, global AI companies may face new sourcing constraints. The case sets a benchmark for how Japanese IP is treated in the AI era.

OpenAI’s response and policy update

In early October 2025, OpenAI’s CEO acknowledged the backlash and committed to more granular controls, signalling a shift toward opt-in permission for copyrighted content and to honour creative rights-holders more proactively.

However, CODA says these changes are not sufficient and remain retrospective rather than preventive. The negotiations are ongoing, and how OpenAI implements the policy will be watched closely.

Business risk for OpenAI

If the dispute leads to legal action, OpenAI could face potential liability, injunctions, or demands for licensing of training data, which would be costly and complex for large scale model training.

Furthermore, a precedent requiring opt-in datasets could raise costs for all large-scale model-training efforts. For OpenAI, the reputational and operational risks are high.

Implications for the broader AI industry

The challenge from major IP holders such as Studio Ghibli and Bandai Namco may prompt industry-wide changes: more opt-in training frameworks, licensing marketplaces for data, or stricter regulation of model outputs that resemble protected content.

Other generative-AI companies will watch this dispute closely for a signal of how copyright law is applied. The outcome may reshape how models are trained, governed, and monetised.

What rights-holders should monitor

Studios and creators should track how AI models are trained, identify the use of their IP in training datasets, evaluate output similarity to their works, and consider formal engagement with AI companies.

They may also assess whether models’ opt-out systems provide adequate protection under their jurisdiction’s law. Documentation and assertions of style, character likeness, or distinct creative elements may become key evidence in disputes.

What users and creators should be aware of

Users of models like Sora 2 should understand that generating other people’s characters or mimicking animation styles, even if technically allowed under the interface, can raise copyright issues for both users and the platform.

Creators should consider how their work can be referenced or licensed within AI ecosystems. The era of “unlimited style mimicry” is facing legal scrutiny, and users may need to rethink what’s permissible.

Thinking about making Ghibli-style AI art? See to learn if Ghibli-style AI art is breaking the law.

Final thoughts

The dispute between Studio Ghibli, Bandai Namco, and OpenAI over Sora 2’s training data is a critical moment for AI, creative rights, and global IP law. For AI companies: reassess data sourcing and consent mechanisms.

For rights-holders: proactively establish mechanisms to engage with model-developers. For users and creators: stay informed about output rights and ethical model usage. The future of generative AI may hinge on how this conflict is resolved.

What will Sony’s $464M deal mean for gamers? See Sony invests $464 million in Bandai Namco to deepen gaming ties.

Do you believe AI models like Sora 2 should require prior permission to use creative works for training, even if the end output is original, and why? Share your thoughts.

Read More From This Brand:

• Sora Video Generator Coming to ChatGPT
• OpenAI rolls out the Sora app and Sora 2 model that might compete with TikTok
• Meet Sora, OpenAI’s Visionary Video Model

Don’t forget to follow us for more exclusive content right here on MSN.

If you like this story, you’ll LOVE our Free email newsletter. Join today and be the first to receive stories like these.

This slideshow was made with AI assistance and human editing.