Eliyahu Gasson | editor-in-chief

Technology continues to evolve. Telephones, the internet, social media, blockchain and now generative AI have both helped solve and create new challenges for us.  

Something good can come from advancement in AI. The videos can be funny, everyone wants to see themselves as a cartoon and writing college essays has been a lot easier, even if the quality is sub par. But bad actors have demonstrated how these tools can be used to lie, slander, cheat, scam and make us dumber. They threaten the careers of artists and human creativity as a whole. They are destroying the environment and are exacerbating human-caused climate change. They pose a direct threat to our lives. It’s time to legislate.

Consider this a part three to my series of articles criticizing generative AI.

OpenAI, the company behind the large language model, ChatGPT, announced the roll out of a new and undoubtedly revolutionary product, Sora 2, giving the average (and perhaps untalented) person the ability to create convincing videos where things that have never nor could never happen, happen.

Videos of dogs driving cars being pulled over by police and internet influencer Jake Paul coming out as gay and applying makeup to his face have begun populating timelines across social media, examples of how this technology will shape our future shared understanding of truth.

However, this concern is not new. A similarly dangerous technology arose near the end of the last decade with deepfakes. In 2018, FakeApp and other software gave the average user the chance to make someone else say something they’ve never said using a photo of their face. Paired with voice cloning software, malicious actors demonstrated why we should not be given access to the fancy new technology. Users have resurrected people from the dead and, because they could, created adult films featuring famous women, most notably pop star Taylor Swift in 2024.

The latter incident sparked outrage among several advocacy groups, politicians and fans of the singer-songwriter.

Just a year later, Swift was the victim of a similar incident when Grok, Elon Musk’s AI agent, generated sexually explicit material featuring her.

While Grok and other AI agents were alright at generating videos, they weren’t perfect. Sometimes the AIs would mess up on hands or eyes. What makes Sora 2 so much more concerning is how good it is at generating convincing videos.

Sora 2 has comedic timing. It can quickly change camera angles and simulate background noise. The characters in the videos talk like real people. They stutter and pause as though they’re speaking naturally.

The team at OpenAI has somehow found a way to mimic the style of the average TikTok incredibly well.

They’ve managed to pull off something truly innovative, but at the same time, incredibly risky.

To OpenAI’s credit, Sora 2 contains safeguards to prevent people from appearing in videos without explicit consent. The company claims that their software will “place you in control of your likeness end-to-end,” according to a news release published on Sept 30. “We also apply extra safety guardrails to any video with a cameo, and you can even set preferences for how your cameo behaves,” explaining how users can control their appearance in other users’ videos.

OpenAI also places a visible watermark on videos generated by Sora 2 to indicate they were generated with their platform, though people have found a way to remove those watermarks with … AI. As a backup,

OpenAI claims their videos contain a metadata signature and maintain reverse-image and audio search tools that can link videos back to Sora 2. This is a necessary safeguard, but requires a level of tech savvy-ness that most internet users may not possess, bringing us back to the main problem.

If we can’t trust what we see online without a Google search at the bare minimum, we shouldn’t expect there to be a shared understanding of truth and basis in fact. Without fact-checkers working around the clock to verify every single video on the internet, we can expect more people believing lies and more harmful content being produced.

President Donald Trump has never been one to slander his political enemies. So it’s no surprise he has been happy to use AI generated content to mock his political foes. A day after the government shut down last week, the president uploaded a phony video to Truth Social in which he throws a red “Trump 2028” hat onto House minority leader Hakeem Jeffries’ head. Another video posted by Trump features the Director of the U.S. Office of Management and Budget Russell Vought as the grim reaper over a parody of Blue Oyster Cult’s “Don’t Fear the Reaper,” which besmirches Democratic politicians and threatens to cut of federal jobs as a result of the shutdown.

OpenAI has kept Sora 2 “invite only,” meaning only people with special codes can access it, since its announcement as they work out kinks. Hopefully, by the time it’s fully accessible, we’ll have some sort of regulation surrounding its use or, at the very least, have found some way to adequately counter the harms from it.

Eliyahu Gasson can be reached at gassone@duq.edu