Sora 2 launches as a social video app

OpenAI launched Sora 2 alongside a standalone iOS social app, positioning it against TikTok and YouTube Shorts rather than as a standalone generation tool. The app produces short AI videos to an invite-only feed, initially in the US and Canada, and centres on a “cameo” feature: a user records themselves once, and can then grant friends permission to insert their likeness into AI-generated clips, with the ability to revoke that permission or delete resulting videos at any time.

The launch’s copyright policy proved the more consequential decision. OpenAI applied an asymmetric standard — explicit consent was required to use a real person’s face or voice, but copyrighted characters could appear in generated videos by default unless the rights holder opted out. Within hours, users had flooded the feed with clips featuring characters such as SpongeBob and Mario, and with generated depictions of deceased public figures. Studios, unions and rights holders objected publicly, and OpenAI reversed the policy within roughly 72 hours, moving to an opt-in model for copyrighted characters with revenue-sharing for rights holders who chose to participate.

The reversal echoed a now-familiar pattern in the generative-media dispute: labs shipping a permissive default to seed adoption, then retreating under rights-holder pressure once misuse became visible and public. It also underlined that the underlying capability — highly realistic, brief, sound-synced video generated from text prompts — had reached a point where the harder problem was no longer the model but the governance of what it was allowed to depict, and of whom.