Japanese Industry Group Demands OpenAI Cease Unauthorized Content Use for AI Training

The Content Overseas Distribution Association (CODA), a prominent Japanese organization dedicated to protecting intellectual property in the entertainment sector, has formally requested that OpenAI discontinue the unauthorized use of its members' copyrighted material for training the Sora 2 video generation tool. This move comes amid growing concerns that OpenAI's advanced text-to-video AI is leveraging Japanese creative works without proper consent, potentially infringing on intellectual property rights.

This development was perhaps inevitable, given the capabilities of Sora 2. Shortly after its release, the technology was noted for its ability to produce AI-generated videos with a distinct Studio Ghibli-esque aesthetic. This capability was highlighted by a viral video showing OpenAI CEO Sam Altman alongside frolicking Pokémon, a scene that playfully yet pointedly raised questions about intellectual property and creative ownership, with Altman himself quipping about potential legal action from Nintendo.

CODA, established in 2002 to combat piracy and champion Japanese intellectual property globally across various media—including video games, film, music, and animation—has taken a firm stance. While Nintendo is not a CODA member, Studio Ghibli, whose distinctive style is often mirrored by Sora 2’s output, is. CODA’s official notice to OpenAI emphasizes that many Sora 2-generated videos bear a strong resemblance to Japanese content, suggesting that these works were used as machine learning data.

The association further argues that if Sora 2 produces content that reproduces or closely mimics specific copyrighted works, the act of replication during the AI's machine learning process could constitute copyright infringement. Although no legal proceedings have been initiated yet, CODA has indicated its readiness to pursue such actions if necessary. They have also clarified that Japan's copyright laws generally require prior permission for the use of copyrighted works, contrasting with any opt-out systems OpenAI might propose.

In response to earlier concerns, Altman had previously stated that OpenAI would grant rights holders more detailed control over character generation within Sora 2, similar to an opt-in model. He also acknowledged the significant creative contributions from Japan and the deep engagement users have with Japanese content. CODA's demand includes two key points: ensuring that member content is not used for AI training without explicit permission, and that OpenAI addresses all claims and inquiries regarding potential copyright infringement by Sora 2's outputs. OpenAI has yet to issue an official response to CODA's formal request.

The unfolding situation underscores the complex challenges arising at the intersection of rapidly advancing AI technology and established intellectual property laws. As AI models become more sophisticated in generating creative content, the debate over how these systems are trained and how original works are protected is intensifying, especially when AI outputs closely mirror existing copyrighted material. This case highlights the need for clear guidelines and potentially new legal frameworks to navigate the evolving landscape of AI-generated content and copyright.