CODA asks OpenAI to stop using CODA members' content to train Sora 2

A coalition of Japan’s major entertainment companies has asked OpenAI to stop using their works to train its Sora 2 video-generation model, saying outputs from the system closely resemble copyrighted anime, manga and game material and that such use was not authorized. The request, delivered last week by the Content Overseas Distribution Association, represents publishers and studios including Aniplex, Bandai Namco, Studio Ghibli, Square Enix, Kadokawa and Shueisha and seeks an immediate end to the use of CODA members’ content for machine learning without prior permission.

CODA said in its letter that it has confirmed a substantial portion of Sora 2 outputs closely resemble Japanese works, and that the replication of specific copyrighted material during model training may constitute infringement. The association also challenged Sora 2’s opt-out approach, under which copyrighted works remain available for training unless rights holders request removal, arguing that under Japan’s copyright system prior permission is generally required and that subsequent objections do not remove liability.

The trade group asked OpenAI to stop using CODA members’ content for model training and to...