Krux
March 30, 2026
OpenAI Blocks Living Public Figures From Sora Videos
Published: March 30, 2026 at 1:49 AM
Updated: March 30, 2026 at 1:49 AM
100-word summary
OpenAI just published safety rules for Sora 2 that ban videos of living public figures unless the person themselves uploads a consenting photo through a "character" feature. Every Sora video now gets visible watermarks and hidden metadata that lets OpenAI trace outputs back to the tool. The company is also scanning audio transcripts to block artists' voice imitations and honoring takedown requests. Teen accounts face stricter filters, and parents can manage what kids see in the Sora feed. The approach reveals an awkward truth: the best deepfake defense might be requiring celebrities to authenticate themselves before AI can render them.
What happened
OpenAI just published safety rules for Sora 2 that ban videos of living public figures unless the person themselves uploads a consenting photo through a "character" feature. Every Sora video now gets visible watermarks and hidden metadata that lets OpenAI trace outputs back to the tool. The company is also scanning audio transcripts to block artists' voice imitations and honoring takedown requests. Teen accounts face stricter filters, and parents can manage what kids see in the Sora feed.
Why it matters
The approach reveals an awkward truth: the best deepfake defense might be requiring celebrities to authenticate themselves before AI can render them.