Turns out there are downsides to feeding your likeness into an AI-powered text-to-video generator. One totally predictable drawback is the userbase wrestling away control of your likeness to create deepfakes of your digital double doing crimes.
OpenAI CEO Sam Altman’s likeness is all over the announcement video for Sora 2’s launch. The trailer in question was itself made entirely in Sora 2, situating Altman in an AI-generated studio, an impossible neon-coloured forest, and even an oversized duck racing event. However, the most popular AI-generated clip on the Sora app right now is reportedly a fabricated surveillance video showing Sam Altman stealing graphics cards, according to OpenAI researcher Gabriel Petersson.
The latest Sora model launched today alongside a dedicated app, with its key social pitch being features that allow you to remix your friends’ creations and upload your own likeness into AI-generated videos. Altman claims there are in-app mitigations against the misuse of someone’s likeness. However, I think the clip where his digital double can be heard saying, “Please, I really need this for Sora inference—this video is too good,” as he clutches a graphics card speaks for itself.
Besides the fresh addition of AI-generated audio, Sora 2 demonstrates clear improvements over the first model released last year. That said, one can take some comfort in the fact it still struggles to accurately render text. Besides the wibbly characters on the security guard’s shirt, Sam Altman’s digital double is seen swiping inventory from a display labelled ‘gratics cards’.
The concept of Altman stealing graphics cards is pretty funny, if only because it mimics what’s happening on an industrial scale to support OpenAI’s real-world datacentres. There are 10 gigawatts of Nvidia chips heading to OpenAI in the near future, albeit bought and paid for by OpenAI’s investors, which also included Nvidia. How’s that work?!
i have the most liked video on sora 2 right now, i will be enjoying this short moment while it lastscctv footage of sam stealing gpus at target for sora inference pic.twitter.com/B86qzUGlMqSeptember 30, 2025
Setting deepfakes of criminal behaviour to the side, it’s also just a wee bit concerning that OpenAI chooses to showcase Sora 2’s capabilities with AI-generated depictions of equally ridiculous and dangerous stunts. No, I’m not trying to stoke a moral panic over that one clip where a guy is shown riding two horses at once before falling on his face and rolling away apparently no worse for wear like a cartoon character.
It’s more the clip of the guy doing a backflip on a board in open water that I’m concerned a younger viewer may attempt to emulate. The Sora app did launch with parental controls via ChatGPT. These tools will allow parents to opt their teen’s account “into a non-personalized feed, [choose] whether their teen can send and receive direct messages, and [offer] the ability to control whether there is an uninterrupted feed of content while scrolling.”
However, Sora 2 clips don’t feature any watermarks to indicate they are AI-generated; there’s no disclaimer to alert, say, a younger viewer that though these clips look realistic, they are not at all based in reality. With a recent Microsoft study suggesting that folks struggle to identify AI-generated still images 62% of the time, I’m concerned about how Sora 2’s clips could spread disinformation or even be leveraged to harass and bully.
From PCGamer latest via this RSS feed