OpenAI's New Sora App Lets You Deepfake Yourself for Entertainment (www.wired.com)

OpenAI launched Sora, an iOS invite-only video app powered by its Sora 2 video-generation model that serves a TikTok-like “For You” feed of short, AI-created clips. It’s the company’s first product that synthesizes both visuals and audio, letting users create a digital likeness by speaking numbers and turning their head; that likeness can be shared as a public cameo or restricted to approved people. Users tap faces to add “cameos,” type a simple prompt (e.g., “fight in the office”), and Sora generates a roughly nine-second clip that includes script, voice, and motion. Early outputs can be impressively convincing (and sometimes funny or eerie), though reviewers also found visual and audio rough edges. The release matters because it normalizes personalized deepfakes as entertainment while exposing real safety, privacy and policy challenges. OpenAI built guardrails—blocking sexual/graphic/self-harm content, certain impersonations of public figures, and other classes of harmful material—and notifies users when their likeness is used, but moderation will be strained as adoption grows. For the AI/ML community, Sora highlights advances in multimodal generation (tight audio-visual coherence and identity consistency) and sharpens debates about consent, provenance, watermarking, copyright opt-outs, and how to detect or limit misuse of synthetic likenesses at scale.