Sora 2 launches with safety built in

Sora 2 arrives as OpenAI extends its generative AI portfolio from text and images to video, a domain where realism and motion amplify potential misuse. The company emphasizes safety as a foundation, not an afterthought, integrating protections like C2PA metadata, content filtering, and strict guardrails for image-to-video with people, especially minors. This reflects an evolution from earlier systems, as the red teaming and tightened policies mentioned build on lessons from ChatGPT image generation and Sora 1.

The characters feature is a notable shift, giving users control over their likeness and consent over usage, which directly addresses deepfake concerns. Similarly, the ban on imitating living artists or existing works in audio generation tackles copyright and identity issues. These measures aim to build trust, which is critical for a social platform where user-generated content is shared.

The product's availability through April 26, 2026, suggests a limited rollout, likely to gather feedback and refine safety. The ultimate test is whether these layered defenses—automated scanning, human review, and user controls—can keep the feed safe without stifling creativity. If successful, Sora could set a standard for safe video generation; if not, it may face backlash as seen in other AI launches. The balance hinges on how well the guardrails handle edge cases and evolving misuse.