OpenAI is rolling out its Sora 2 model and accompanying Sora app with a robust safety framework built in from the ground up. The company emphasizes concrete protections to ensure responsible AI video generation.
Every Sora output will carry both visible and invisible provenance signals. This includes C2PA metadata, an industry-standard signature, and internal tools capable of tracing videos back to Sora with high accuracy. Many videos will also feature dynamic, visible watermarks identifying the creator.
Personal Likenesses and Consent
The introduction of image-to-video generation with real person likenesses comes with stringent safety measures. Users must attest to having consent from individuals featured in uploaded photos and possess the necessary rights to the media. These safeguards are even more rigorous than those applied to Sora Characters, formerly the cameo feature.
Images featuring children or those who appear young are subjected to heightened moderation and stricter guardrails. Videos created from such images will always include watermarks upon sharing.
OpenAI is also enabling consent-based likenesses using characters. Users have strong control over how their appearance and voice are used, with guardrails designed to ensure consent. Only the user dictates who can utilize their characters, and access can be revoked at any time. Public figures are generally blocked, except when using the character feature.
Creators can review, delete, and report any videos featuring their characters, ensuring transparency and control. Additional safety guardrails are applied to all character-based videos, with an option for users to enable even stricter settings. These can limit major appearance alterations, prevent embarrassing situations, and maintain identity consistency.
Teen Safeguards and Content Filtering
Sora incorporates enhanced protections for younger users, including limitations on mature content. The feed is designed to be age-appropriate, with harmful, unsafe, or age-inappropriate material filtered from teen accounts. Teen profiles are not recommended to adults, and direct messaging between adults and teens is blocked.
Parental controls within the broader ChatGPT ecosystem allow guardians to manage teen messaging and select a non-personalized feed within the Sora app. Teens also face default limits on continuous scrolling.
The platform employs layered defenses to maintain a safe feed while allowing creative expression. At the creation stage, guardrails aim to block unsafe content, such as sexual material, terrorist propaganda, and self-harm promotion, by analyzing prompts and outputs across video frames and audio transcripts. OpenAI has conducted red-teaming exercises to identify novel risks, tightening policies based on Sora's enhanced realism and the addition of motion and audio compared to earlier AI Models: ChatGPT, Claude, Gemini, and Beyond.
Automated systems continuously scan all feed content against Global Usage Policies, filtering out unsafe or age-inappropriate material. These systems are updated based on emerging risks and are supplemented by human review focused on high-impact harms.
Audio and User Recourse
The addition of audio introduces new safety considerations. Sora automatically scans generated speech transcripts for policy violations and blocks attempts to imitate living artists or existing musical works. Systems are designed to detect and halt such prompts, and takedown requests from creators alleging infringement are honored.
Users retain control over sharing and can remove published content at any time. Videos are only shared to the feed upon explicit user choice. Every video, profile, direct message, comment, and character can be reported for abuse, with clear recourse mechanisms in place for policy violations. Users can also block accounts to prevent profile viewing, character usage, and direct messaging.
OpenAI’s approach underscores a commitment to mitigating risks associated with advanced AI video generation, as detailed in their ongoing research, including their OpenAI Details Malicious AI Use in 2026 efforts.
