OpenAI launches Sora, AI‑generated short‑video social app
OpenAI released Sora on Oct. 3, 2025, a vertical short‑video social app that generates 10‑second videos from text prompts; NPR’s early review found the app can produce highly realistic videos — including deepfake depictions of real people and simulations of violent events — and flagged guardrail gaps despite company‑published policies and in‑app controls. Researchers warn the app may accelerate misinformation and pose security risks, while OpenAI says downloaded videos carry moving watermarks, embedded metadata, and user controls over whether their likeness may be reused.
AI & Tech
National security
🔍 Key Facts
- Release: OpenAI launched the Sora (Sora 2) app on Oct. 3, 2025 as a short‑form AI video social platform.
- Company controls: OpenAI publishes a Sora system card, says videos include moving watermarks and embedded metadata and offers users 'end‑to‑end' face‑use controls (everyone / friends / only me) and takedown options.
- NPR findings: In NPR’s hands‑on testing, Sora generated realistic videos that included fabricated historical footage (e.g., a Nixon 'moon‑landing‑fake' clip), a fake Neil Armstrong helmet removal, a simulated drone attack on a power plant, and short clips touching on chemical/biological/radiological/nuclear topics — demonstrating both deepfake quality and policy‑enforcement gaps.