Studio Ghibli and others stand against OpenAI to protect Japanese culture! "Respect and Systems" for Generative AI — A Game-Changer Approaching for Japanese IP

The "Japan-Origin AI Shock" Surrounding "Sora 2" — Two Demands CODA Presented to OpenAI and Their Ripple Effects

In the early hours of Tuesday, November 4, 2025, Japan time, it was widely reported by overseas tech media that the Japanese industry group CODA (Content Overseas Distribution Association), which includes Studio Ghibli, Bandai Namco, and Square Enix, officially presented two demands to OpenAI regarding its video generation app "Sora 2": "Do not use the copyrighted works of member companies for learning" and "Respond sincerely to claims of rights infringement." The issue began when, after Sora 2 was released on September 30 (UTC), videos resembling Japanese characters and anime expressions spread widely on social media.The Verge

On October 27, CODA submitted a document to OpenAI (made public on October 28). The document clearly states that "the act of reproduction in the process of machine learning itself can constitute infringement," and expressed the view that operations based on "post-facto opt-out" from rights holders contradict the principles of Japanese law (prior consent).coda-cj.jp

In mid-October, the Japanese government also reportedly made a formal request to OpenAI to stop infringing on manga and anime, which are considered "Japan's treasures," with figures like Minoru Kihara, who is not the Kihara referred to in the article, but rather Michihiro Kounai (Minister of State for Special Missions, etc.), involved. Criticism stems from the fact that Mario and Pikachu-like videos were "unchecked" on Sora.The Verge

From Opt-Out to Opt-In? OpenAI's "Policy Shift" and the Prospect of Licensing

In early October, OpenAI announced plans for "official fictional character cameos" in Sora. Additionally, they indicated a review of Sora's copyright handling from an "opt-out-centric" approach, aiming to offer rights holders more detailed control and revenue-sharing mechanisms. This effectively signals a shift towards an opt-in model (prior consent and licensing).The Verge

Behind this "policy shift" is the "Ghibli-style" image movement that spread shortly after the release of GPT-4o. At that time, OpenAI reportedly took a conservative approach by limiting the specification of individual artist styles and gradually narrowing the scope of broader "artistic styles" such as studio names.Business Insider

Point of Contention: Is Reproduction During Learning "Infringement" or Cultural Reference?

The principle of Japanese law is "prior consent." CODA argues that the act of reproducing copyrighted works for learning purposes itself can constitute infringement. On the other hand, the generative AI side often counters with "it's statistical learning and not reproducing the original work" and "the output is transformation and alteration." The recognition of the fact that outputs "resembling existing characters" proliferated after the release of Sora 2 will significantly influence future discussions.coda-cj.jp

The Atmosphere on Social Media: Advocacy, Opposition, and Realistic "Distribution" Discussions

There are three major trends visible on social media.

1. Caution from Creators/Fans
   Comments such as "They're freeloading on the lines and colors of my favorite artists" and "Threatening the livelihood of artists" highlight ethical and real damage concerns. This sentiment has been strong since the "Ghiblification" boom in spring. Major media have also reported on the impact of "Ghiblification" on copyright and creative professions, capturing voices of opposition.AP News

2. Rebuttals and Enthusiasm from AI Advocates
   On platforms like Reddit and X, reactions such as "Sora 2 is a paradigm shift" and "Don't halt evolution with regulations" remain strong. However, there are also instances of "riding the wave" by falsely claiming fake videos as Sora 2 creations, which is said to be amplifying confusion.Reddit

3. Expectations for Licensing as a Realistic Solution
   Posts welcoming "official cameos and revenue sharing" and the emergence of a new market for "fan fiction x official IP" have increased. Overseas and domestic news and analyses are boosting interest in the opt-in model and revenue-sharing design.The Verge

Elsewhere, Japanese tech media and curation platforms repeatedly discuss the "clash of culture and law" brought about by Sora 2, with political, legal, and industry players increasingly reacting.窓の杜

A Retrospective Schedule of "Sora 2 x Japanese IP"

• September 30, 2025 (UTC): Sora 2 released. Videos resembling Japanese IP spread.The Verge

• October 15, 2025 (UTC): Japanese government reportedly requests OpenAI for corrections.The Verge

• October 27-28, 2025 (JST): CODA submits and publishes a document to OpenAI.coda-cj.jp

• Early October 2025: OpenAI hints at "cameo" plans and revenue-sharing consideration.The Verge

• November 3, 2025 (UTC): The Verge reports a summary of the issue (Japan time November 4).The Verge

What Might Happen Next

• Intensification of License Negotiations: Especially for Japanese IP, brand management is strict. "Official appearances" on Sora will require detailed condition setting, including supervision, certification, and withdrawal rights by rights holders.The Verge

• Legal Test Cases Surrounding "Learning Reproduction": If the first practical judgment based on Japanese law is made, it could shake the global design principles of AI learning.coda-cj.jp

• Reorganization of Platform Design: Strengthening governance "before and after generation" is unavoidable, including style specification and prompt restrictions, copyright filtering and detection, and automation of revenue sharing.Business Insider

Conclusion: "Respect" for Culture, "System" for Technology

The CODA letter this time is an attempt to clearly redraw the boundary between "reference" and "plagiarism" in the AI era from the standpoint of Japanese law and culture. The "official cameo x revenue sharing" envisioned by OpenAI could be one point of compromise, but to make it truly sustainable, a comprehensive design including prior consent and transparency from the learning stage, and strong detection and withdrawal flows at the output stage, will be necessary.coda-cj.jp