Safe Creation with Sora

Sora 2 and the Sora app offer state-of-the-art video generation capabilities, and strong safety measures were built in from the design stage.

To identify AI-generated content, they include C2PA metadata and visible and invisible provenance signals. They also use internal reverse image and audio search tools to trace whether content was generated by Sora with high accuracy, and apply dynamic watermarks that include the creator's name.

When generating videos that use the likeness of real people, the following strict standards apply:

• Image-to-video: Consent and rights regarding the person are confirmed from the user, and content involving children and teens in particular undergoes even stronger moderation.
• Characters feature: To protect a user's appearance and voice (Likeness), it can only be used with the person's own consent, and depictions of public figures are blocked. Users can review and delete videos that use their character at any time.

For teen protection, inappropriate content is filtered, and the system is designed so that teen profiles are not recommended to adults. Parental controls are also provided so parents can manage whether messages are received or configure feed settings through ChatGPT.

To prevent audio and harmful content, prompts and outputs are inspected from multiple angles, and attempts to imitate the voice or music of real artists are blocked. Users can delete content they have posted at any time and decide for themselves whether to share it.