Sora 2: Security and Innovation at the Heart of Video Creation

Sora 2: A Secure Advancement in Video Creation

The Sora 2 model, paired with the Sora application, offers cutting-edge video generation technology while integrating security measures from the ground up. The goal is to enable safe collaborative creation, thanks to well-defined protections.

Traceability of AI-Generated Content

Every video produced with Sora is equipped with provenance signals, both visible and invisible. These videos incorporate C2PA metadata, an industry-recognized standard, which allows for precise tracking of videos back to their source. Internal tools for reverse image and audio searches, inspired by the systems used for ChatGPT and Sora 1, enhance this traceability. Additionally, many videos display dynamic watermarks, including the creator's name, for clear identification.

Video Creation from Real Images

Sora allows users to transform photos of family and friends into videos, provided that the user has obtained consent from the individuals involved. Videos featuring real people are subject to very strict safeguards, surpassing those of Sora Characters. Images of children or individuals who appear young are particularly monitored, with increased restrictions on their use. These videos are systematically marked with watermarks when shared.

Control Over Character Likeness

Sora provides the ability to create characters, allowing users to control their appearance and voice within the application. Safeguards are in place to ensure that the use of these likenesses is done with the user's consent. Users can decide who can use their characters and revoke that access at any time. We also take measures to block representations of public figures, except for those using the character feature. Videos created with these characters, including drafts from other users, remain visible to the concerned user, who can delete or report them if necessary. Additional safeguards can be activated to restrict major changes to appearance or embarrassing situations.

Protection Measures for Teens

Sora has implemented enhanced protections for young users, limiting access to mature content. The feed is designed to be age-appropriate, and potentially harmful or inappropriate content is filtered for teen accounts. Teen profiles are not visible to adults, who cannot initiate conversations with them. By default, teens also have limits on the amount of continuous scrolling in Sora. Parents can manage their children's interactions via ChatGPT, controlling direct messages and choosing a non-personalized feed.

Filtering Harmful Content

Sora employs a layered approach to secure the feed while encouraging creativity. Safeguards block dangerous content before it is created, including sexual content, terrorist propaganda, or incitement to self-harm. A dedicated team tests for new risks, and image generation policies are strengthened to account for Sora's increased realism. Automated systems analyze the feed content to detect and filter dangerous or inappropriate materials, supplemented by human review. The systems are continuously updated as we learn about new risks.

Enhanced Audio Security

Adding audio in Sora requires additional security measures. While perfection is difficult to achieve, Sora invests in automatic transcription analysis to detect policy violations. Attempts to create music imitating living artists or existing works are blocked. The systems are designed to identify and stop these initiatives, and requests for removal from creators who believe their work has been infringed are honored.

User Control and Recourse

Users have complete control over sharing their videos, able to delete them at any time. Videos are only shared in the feed at the user's discretion. Every piece of content, whether videos, profiles, direct messages, comments, or characters, can be reported for abuse. Clear recourse options are available in case of policy violations. Users can also block accounts, preventing any unwanted contact or use of their character.