Sora 2 от OpenAI: Виртуальный художник дезинформации создает убедительные дипфейки Translation: Sora 2 by OpenAI: The Virtual Artist of Disinformation Creates Convincing Deepfakes

The Sora 2 video model by OpenAI generated realistic videos with misleading claims in 80% of instances upon receiving a relevant request, according to a study conducted by experts from NewsGuard.

Out of 20 prompts, 16 successfully resulted in the creation of disinformation, with 11 produced on the first attempt. The authors highlighted that five topics had previously been used in Russian disinformation campaigns.

The application generated deceptive footage, including a supposedly Moldovan official destroying pro-Russian ballots, U.S. immigration services detaining a small child, and a Coca-Cola representative stating that the company would not sponsor the Super Bowl.

None of the aforementioned incidents actually occurred. The videos appeared sufficiently credible to mislead a user casually scrolling through their feed.

NewsGuard experts discovered that creating such material takes just a few minutes and requires no technical expertise, while watermarks can be easily removed.

«Some videos produced using Sora appeared more convincing than the original posts that spawned viral falsehoods. For instance, a video featuring the arrest of a child by agents from the ICE looked far more realistic than the initial blurry and cropped image that initiated the false claim,» the research stated.

OpenAI encountered another issue after the release of Sora 2 — deepfakes involving Martin Luther King Jr. and other historical figures. Users created videos showing the civil rights leader allegedly shoplifting, evading police, and embodying racial stereotypes. His daughter described such videos as «demeaning» and «absurd.»

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

A similar situation arose with dozens of other well-known individuals. Robin Williams’ daughter, Zelda, requested via Instagram that people refrain from sending her AI-generated videos featuring her father.

Within three weeks, the company revised its policies multiple times: initially allowing fake videos, then instituting a consent system for rights holders, later blocking the use of certain figures, and ultimately implementing affirmation rules from celebrities and voice protections.

It is worth noting that in October, deepfakes featuring Sam Altman overwhelmed OpenAI’s new social application Sora.