Kiss reality goodbye: AI-generated social media has arrived (www.npr.org)

OpenAI released Sora (Sora 2 in NPR’s preview), a short-form, vertical-video app that generates addictive 10-second clips from simple text prompts—think TikTok populated entirely by AI-made videos. The app provides user controls for “end-to-end” face usage (allowing users to permit or block others from using their likeness), moving Sora watermarks and embedded metadata to label AI-origin, plus in-app reporting and a mix of automated and human review. OpenAI says it prohibits content for deception, fraud, impersonation and other harms and will honor takedown requests from rights holders. But early testing by NPR and researchers showed the technology’s power and the enforcement gaps: Sora can produce disturbingly realistic deepfakes (e.g., fabricated Nixon and Armstrong footage), violent scenes like drone strikes, and even CBRN‑adjacent content despite policy limits. The app is already saturated with trademarked and copyrighted characters, which OpenAI currently allows but will block on request. The result is a fast path to democratized video synthesis that raises familiar but amplified problems—misinformation, impersonation, copyright risk and moderation at scale—signaling a possible era where “seeing is not believing” and forcing urgent technical, legal and policy responses from the AI/ML community.