Are Sora 2 and other AI video tools risky to use? Here's what a legal scholar says (www.zdnet.com)

OpenAI’s Sora 2 — a consumer-facing AI video tool that can remix people, voices and scenes with striking fidelity — exploded in popularity and immediately provoked real-world worry. A ZDNET reporter demonstrated how easy it was to generate a fake endorsement of OpenAI CEO Sam Altman in minutes, prompting backlash from Hollywood rights holders and a public rebuke from the Motion Picture Association. OpenAI has since added guardrails (a Sora 2 System Card and “feed philosophy”), consent-based “cameo” controls, IP and audio imitator blocks, takedown/reporting workflows and prompt rejections for flagged likenesses — but rights holders remain skeptical about enforcement and responsibility. Legally and technically the Sora 2 story crystallizes several tensions for the AI/ML community: Yale’s Sean O’Brien says courts are trending toward a four-part reality — copyright protects only human-created works, many AI outputs may be uncopyrightable, the human operator is liable for infringing outputs, and training on copyrighted data without permission is actionable. Creatively, Sora 2 democratizes skills (lowering barriers for novices) while threatening livelihoods and raising questions about provenance, attribution and theft when models are trained on existing art. The takeaway: high-fidelity generative video shifts the problem from “can we build it?” to “how do we assign liability, enforce rights, and design robust technical and policy controls (watermarks, provenance, stricter filters) before misuse proliferates?”