Sora 2 от OpenAI: Путь к созданию правдоподобных фейков за считанные минуты Translation: Sora 2 by OpenAI: The Path to Crafting Convincing Fakes in Minutes

The video model Sora 2 from OpenAI generated realistic videos containing false claims 80% of the time after receiving appropriate prompts. The research was conducted by experts from NewsGuard.

Out of 20 prompts, 16 successfully resulted in the production of misinformation, with 11 being created on the first attempt. The authors noted that five of these topics had previously been employed in Russian campaigns to spread fake news.

The application produced fake footage depicting a Moldovan official supposedly destroying pro-Russian ballots, a U.S. immigration service detaining a small child, and a Coca-Cola representative declaring that the company would not sponsor the Super Bowl.

None of these events actually occurred. The videos appeared realistic enough to mislead users quickly scrolling through their feeds.

NewsGuard experts found that creating such content takes just a few minutes and does not require any technical expertise, while watermarks can be easily removed.

«Some videos generated by Sora looked more convincing than the original posts that spawned viral hoaxes. For instance, the clip showing the detention of a child by ICE agents is considerably more realistic than the original blurry and cropped image that initiated the false claim,” the study states.

OpenAI faced another issue after releasing Sora 2—deepfakes involving Martin Luther King Jr. and other historical figures. Users created videos in which the civil rights leader supposedly shoplifted, fled from police, and replicated racial stereotypes. His daughter described such videos as «degrading» and «absurd.»

The startup acknowledged that the video generator produced «disrespectful» content and removed the option to use King’s likeness.

A similar situation occurred with dozens of other well-known individuals. Robin Williams’ daughter, Zelda, requested via Instagram not to send her AI-generated videos featuring her father.

Over the course of three weeks, the company revised its policies multiple times: first, it allowed fake videos, then it implemented a consent system for rights holders, later it blocked the use of certain figures, and ultimately it established rules for approval from celebrities and voice protection.

It’s worth mentioning that in October, deepfakes featuring Sam Altman flooded the new social application Sora from OpenAI.