OpenAI launched its new text-to-video model, Sora 2, with a companion social app, but its initial policy requiring copyright holders to "opt-out" of having their IP used—a detail first reported by The Wall Street Journal—sparked an immediate backlash and a rapid course correction from the company.
An IP free-for-all: The app's standout "cameos" feature lets users inject their own likeness into generated scenes, but within hours of launch, the platform was overrun with synthetic videos featuring a rogue's gallery of familiar faces. The feed filled with clips starring figures from "SpongeBob SquarePants" and "Breaking Bad," with one viral creation showing the characters in a meth-lab parody, complete with uncanny, AI-generated voices.
Hollywood strikes back: The backlash from creators was immediate. The Hollywood Reporter revealed that major talent agency WME had already informed OpenAI it was opting out all of its clients. In response, CEO Sam Altman announced the company would walk back its stance, promising rightsholders "more granular control" and floating a potential revenue-sharing model for what he termed "interactive fan fiction."
Disinformation on demand: The copyright chaos was only half the problem. The app also proved to be a powerful tool for generating convincing disinformation, as users created realistic videos depicting crimes, including CCTV footage of Sam Altman shoplifting GPUs and synthetic police bodycam videos.
The Sora 2 launch shows OpenAI is sticking to its "launch fast, fix later" playbook, leaving the fundamental tension between rapid innovation and responsible deployment unresolved.
Beyond the launch-day drama, the push into video is part of OpenAI's larger strategy to train more capable AI models on visual data in its quest for AGI. The Sora controversy is just the latest front in a broader legal war between media companies and AI firms over copyright. And in a strange twist, the model is apparently blurring the lines between creative and cognitive tasks by showing it can also answer science questions in its generated videos.
