Sora app’s hyperreal AI videos ignite online trust crisis as downloads surge

Scrolling through the Sora app can feel like entering a real-life multiverse.

Michael Jackson does stand-up; the alien from the “Predator” films cooks hamburgers at McDonald’s; a home security camera captures a moose crashing through the glass door; Queen Elizabeth dives off a table in a pub.

Such improbable realities, fantastical futures, and absurd videos are the mainstay of the Sora app, a new short-form video app released by ChatGPT creator OpenAI.

The continuous stream of short, hyper-realistic videos made by artificial intelligence is breathtaking and fascinating at first. But it quickly triggers a new need to reconsider every piece of content as real or fake.

“The biggest risk with Sora is that it makes plausible deniability impossible to overcome and erodes confidence in our ability to distinguish the real from the synthetic,” said Sam Gregory, a deepfakes expert and executive director of WITNESS, a human rights organization. “Individual fakes are significant, but the real damage is a fog of doubt that settles over everything we see.”

All videos in the Sora app are completely AI-generated and there is no option to share real footage. But within the first week of its launch, users were sharing their Sora videos on all types of social networks.

Less than a week after its launch on September 30, the Sora application reached a milestone million downloadssurpassing ChatGPT’s initial growth. Sora also reached the top of the App Store in the United States. For now, the Sora app is only available to iOS users in the United States, and people can only access it if they have an invite code.

To use the app, users must scan their face and read three numbers displayed on the screen for the system to capture a voice signature. Once done, users can enter a custom text prompt and create hyper-realistic 10-second videos with background audio and dialogue.

Through a feature called “Cameos,” users can superimpose their face or that of a friend into any existing video. Although all outputs carry a visible watermark, many websites now offer watermark removal for Sora videos.

When it launched, OpenAI took a lax approach to enforcing copyright restrictions and allowed the recreation of copyrighted material by default unless owners opted out.

Users began generating AI videos featuring characters from titles such as “Sponge Bob,” “South Park” and “Breaking Bad,” as well as videos inspired by the game show “The Price Is Right” and the ’90s sitcom “Friends.”

Then came reenactments of deceased celebrities, including Tupac Shakur roaming the streets of Cuba, Hitler confronting Michael Jackson and remixes of the Rev. Martin Luther King Jr. giving his iconic “I Have A Dream” speech — but calling for the release of disgraced rapper Diddy.

“Please stop sending me AI daddy videos,” Zelda Williams, daughter of the late comedian Robin Williams, posted on Instagram. “You’re not making art, you’re making disgusting, over-processed hot dogs out of human lives and art history and music, and then you feed them to someone else, hoping they’ll give you a little push and like it. Disgusting.”

Other reconstructions of deceased celebrities, including Kobe Bryant, Stephen Hawking and President Kennedy, created on Sora, have been posted on social media sites, garnering millions of views.

Christina Gorski, director of communications at Fred Rogers Productions, said Rogers’ family was “frustrated by AI videos misrepresenting Mister Rogers posted online.”

Videos of Mr. Rogers holding a gun, saluting rapper Tupac, and other fake satirical situations were widely shared on Sora.

“The videos are in direct contradiction to the careful intentionality and respect for fundamental principles of child development that Fred Rogers brought to each episode of Mister Rogers’ Neighborhood. We have contacted OpenAI to request that Mister Rogers’ voice and likeness be blocked for use on the Sora platform, and we expect that they and other AI platforms will respect the personal identities in the future,” Gorski said in a statement to The Times.

Hollywood talent agencies and unions, including SAG-AFTRA, have begun accusing OpenAI of inappropriate use of likenesses. The central tension comes down to control over the use of licensed actor and character likenesses – and fair compensation for their use in AI videos.

Following Hollywood’s concerns over copyright, Sam Altman has shared a blog post, promising greater control for rights holders to specify how their characters can be used in AI videos – and exploring ways to share revenue with rights holders.

He also said that studios could now “opt in” to their characters being used in AI recreations, a reversal from OpenAI’s initial stance of an opt-out regime.

The future, according to Altman, is moving toward creating personalized content for an audience of a few – or an audience of one.

“Creativity may be about to experience a Cambrian explosion, and with it, the quality of art and entertainment may increase dramatically,” Altman wrote, calling this kind of engagement “interactive fan fiction.”

The estates of deceased actors, however, are scrambling to protect their image in the age of AI.

CMG Worldwide, which represents the estates of deceased celebrities, has partnered with a deepfake detection company Loti AI to protect CMG cast lists and domains from unauthorized digital use.

Loti AI will continuously monitor impersonations of 20 personalities represented by CMG, including Burt Reynolds, Christopher Reeve, Mark Twain and Rosa Parks.

“Since the launch of Sora 2, for example, our registrations have increased 30-fold, as people look for ways to take back control of their digital image,” said Luke Arrigoni, co-founder and CEO of Loti AI.

Since January, Loti AI said it has removed thousands of instances of unauthorized content as new AI tools have made it easier than ever to create and distribute deepfakes.

After numerous “disrespectful depictions” of Martin Luther King Jr., OpenAI said it was suspending the generation of videos featuring the civil rights icon on Sora, at the request of King’s estate. While there are strong free speech interests in the depiction of historical figures, public figures and their families should ultimately have control over how their likeness is used, OpenAI said in a post.

Now, authorized representatives or landowners can request that their image not be used in Sora cameos.

As legal pressure increases, Sora has become stricter about when it will allow the recreation of copyrighted characters. It is publishing more and more notices of content policy violations.

Now, creating Disney characters or other images triggers a content policy violation warning. Users who are not fans of the restrictions have started creating video memes about content policy violation warnings.

There is growing virality in what has been dubbed “AI slop.”

Last week, circular camera footage of a grandmother chasing a crocodile out the door was shown, along with a series of videos from “big Olympics” where obese people take part in sporting events such as pole vaulting, swimming and track and field.

Dedicated waste plants have turned this commitment into a revenue stream, generating a constant stream of videos that are hard to look away from. One pithy tech commentator dubbed it “Cocomelon for adults.”

Even with increasing protection for celebrity likenesses, critics warn that casual “likeness appropriation” of an ordinary person or situation could sow public confusion, increase misinformation, and erode public trust.

Meanwhile, even though the technology is used by bad actors and even some governments for propaganda purposes and to promote certain political views, those in power can hide behind the flood of fake news by claiming that even real evidence was generated by AI, WITNESS’ Gregory said.

“I am concerned about the possibility of fabricating images of protests, staging fake atrocities or inserting real people with words put into their mouths into compromising scenarios,” he said.