Michael Jackson, Elizabeth II, Saddam Hussein: the AI ​​reappearance of deceased personalities amuses and exasperates

OpenAI’s app, launched in September and widely considered a deepfake machine, has produced a stream of videos of historical figures including Winston Churchill, as well as celebrities such as Michael Jackson and Elvis Presley.

In a TikTok clip reviewed by AFP, Queen Elizabeth II, wearing a pearl necklace and crown, arrives at a wrestling match on a scooter, crosses the rope and jumps over a wrestler. In another clip on Facebook, the queen praises some cheese puffs in a supermarket aisle, while in another she is seen playing soccer.

But not all the videos generated by OpenAI’s Sora 2 model have provoked laughter.

In October, the AI ​​giant prevented users of its Sora app from creating videos using Martin Luther King Jr.’s image following complaints from his estate. Some users had made videos in which this civil rights icon made monkey noises during his famous “I Have a Dream” speech, and uttered rude, offensive or racist comments.

“Maddening”

“We are entering the ‘uncanny valley,’” explained Constance de Saint Laurent, a professor at Maynooth University in Ireland, referring to the theory according to which, when an object reaches a certain degree of anthropomorphic similarity or hyperrealism, it causes a feeling of anguish and discomfort.

LOOK: The most used password in 2025 is still “123456”, despite years of security warnings

“If you suddenly started receiving videos of a deceased loved one, it would be traumatic,” he told AFP. “These videos have real consequences.”

In recent weeks, the children of actor Robin Williams and American activist Malcolm X have condemned the use of Sora to create videos of their deceased parents.

Zelda Williams, daughter of Robin Williams, recently asked Instagram to stop sending her AI-generated clips of her father. It’s “maddening,” he said.

An OpenAI spokesperson told AFP that while “in the interests of freedom of expression there is a strong interest in the representation of historical figures,” public figures and their families should have ultimate control of their image.

For “recently deceased” figures, authorized representatives can now request that their image not be used in Sora, it added.

stop trusting

“Despite OpenAI’s claim to allow people to control their image, they have published a tool that clearly does the opposite,” Hany Farid, co-founder of the American cybersecurity company GetReal Security and professor at the University of California at Berkeley, told AFP.

LOOK: 2026, the year in which space will be looked at again: the Moon and Mars continue to be the main objectives

“While they have (largely) put an end to the creation of Martin Luther King Jr. videos, they do not stop users from appropriating the identities of many other celebrities,” he lamented.

“Beyond OpenAI implementing certain protections” for that political figure, “there will be another AI model that does not do so, and therefore this problem will only get worse,” Farid warned.

This became evident following the murder this month of Hollywood director Rob Reiner, when AFP fact-checkers discovered AI-generated clips of his image spreading online.

As advanced AI tools proliferate, vulnerability is no longer limited to public figures: non-celebrity deceased people can also have their names, likenesses and words distorted for the purposes of manipulation.

Researchers warn that the uncontrolled spread of synthetic content – ​​called “AI slop” – could ultimately drive users away from social networks.

“The problem with misinformation in general is not so much that people believe it. Many people don’t believe it,” Saint Laurent said. “The problem is that they see real information and they don’t trust” that it is true, he noted. And tools like Sora are going to “massively amplify this phenomenon.”