New AI video gen from OpenAI raises new questions (handyai.substack.com)

OpenAI today released Sora 2, a major leap in AI video generation that produces up to 20-second clips with synchronized dialogue, sound effects and markedly improved physics and world consistency. Compared with the first Sora, Sora 2 models buoyancy, rigidity and realistic failure modes (e.g., a failed kickflip behaves like a real failure rather than “teleporting” the skateboard), preserves state across multi-shot prompts, and supports diverse cinematography and stylistic controls. A standout feature, Cameos, lets users register their likeness and voice via a one-time recording so the model can reliably insert that person into generated scenes—access is controllable and revocable by the cameo owner. OpenAI is launching Sora as an invite-only iOS app dedicated exclusively to AI-generated video, arguing separation protects content integrity, creator economics and enables tighter moderation at creation time. The launch also exposes sharp tensions: OpenAI’s opt-out approach to copyrighted training data shifts the burden onto creators to remove their work, inviting ethical and legal pushback and fueling calls for opt-in consent. Technically impressive, Sora 2 accelerates new creative possibilities but intensifies questions about rights, discovery, platform design and the economic impact on human creators—decisions now will shape how synthetic media integrates with society and law.