CODA, a Japanese content trade group whose members include Studio Ghibli, Square Enix, and Bandai, has asked OpenAI to stop using its members’ content to train the company’s Sora 2 model. The letter from CODA is dated October 27 and follows the October 1 launch of Sora 2, OpenAI’s flagship video and audio generation model.
The letter, posted on CODA’s website, says in part: “CODA has confirmed that a large portion of content produced by Sora 2 closely resembles Japanese content or images. CODA has determined that this is the result of using Japanese content as machine learning data.” The group argues that when specific copyrighted works are reproduced or similarly generated as outputs, that replication during the machine learning process may amount to copyright infringement.
CODA takes issue with OpenAI’s opt-out approach for Sora 2, which requires copyright holders to notify OpenAI to have material excluded. The letter says Japan’s copyright system generally requires prior permission for use of copyrighted works and does not allow avoiding liability through later objections. CODA asks that its members’ content not be used for machine learning without permission and that OpenAI respond to claims from CODA member companies about potential copyright infringement.
The group lists two concrete requests:
- That CODA members’ content not be used for machine learning for Sora 2 without prior permission
- That OpenAI respond sincerely to claims and inquiries from CODA member companies regarding alleged copyright infringement tied to Sora 2 outputs
This complaint lands amid wider Japanese concern about AI and copyrighted works; Japanese government officials recently spoke out about the need to protect manga, anime, and music from unauthorized training uses. The wider debate over how online material is harvested for models has led platforms and communities to take action before, such as when Reddit blocked the Internet Archive to limit scraping, readers can find more on that in our coverage of Reddit blocking the Internet Archive to stop AI scraping its content.
OpenAI’s Sora 2 page describes the model and its capabilities in video and audio generation, and CODA’s letter points explicitly to outputs that it says closely mirror Japanese works. The letter asks OpenAI to engage with CODA member companies so that creators’ rights and copyright concerns are addressed alongside AI development.
Join the conversation and follow us on X, Bluesky, and YouTube.
