Executive Summary
OpenAI has outlined its comprehensive safety framework for the new Sora 2 model and application. The approach focuses on proactive measures to prevent misuse by embedding content provenance signals, requiring user consent for creating videos with real-person likenesses, and implementing robust content filtering. Special protections are in place for teen users, and a new "Characters" feature gives individuals explicit control over their own digital likeness, underscoring OpenAI's strategy to balance creative capabilities with responsible AI deployment.
Key Takeaways
* AI Content Provenance: Every video generated by Sora 2 will contain both visible watermarks and invisible C2PA metadata to clearly identify it as AI-created.
* Consent for Real Likeness: Users can create videos from photos of real people after attesting they have consent. These generations are subject to stricter safety guardrails, with even tighter restrictions for content involving minors.
* "Characters" Feature: A new consent-based system allows users to control their own digital likeness (appearance and voice). Users can grant and revoke access for others to use their "character."
* Teen Safeguards: The platform includes enhanced protections for teens, including filtering of age-inappropriate content, preventing adults from initiating messages, and parental controls.
* Harmful Content Filtering: Layered defenses scan prompts, video frames, and audio transcripts to block unsafe content like sexual material, terrorist propaganda, and self-harm promotion.
* Audio Safeguards: The system scans generated speech for policy violations and blocks attempts to create music that imitates living artists or existing works.
* User Control: Users retain full control over sharing their videos and are provided with tools to report content, block other accounts, and manage their privacy.
Strategic Importance
This announcement is a proactive move to address public and regulatory concerns about the misuse of realistic AI video, establishing a framework for responsible deployment. It positions OpenAI as a safety-conscious leader in the generative AI space, aiming to build user trust before a wider release.