Hurricane Melissa: When AI-Generated Fake Videos Distort Information

No, Jamaicans are neither partying nor jet-skiing while waiting for the rain. They are eagerly preparing for the destructive passage of Hurricane Melissa in the Caribbean, amidst a dangerous surge of misinformation generated by AI.

As political authorities and rescue teams organize and disseminate appropriate messages to at-risk populations, AFP has identified dozens of misleading content on social media—most of which bear the watermark of Sora, an OpenAI video generation model. OpenAI did not respond to AFP's inquiries.

The hurricane, which could be the most violent to hit Jamaica and has already resulted in several fatalities, poses a risk of catastrophic flooding and landslides.

On Monday, Jamaican Information Minister and senator Dana Morris Dixon stated that she and other ministers had joined a press conference to provide "accurate information" regarding this monstrous storm.

"I am in many WhatsApp groups, and I see all these videos coming in. Many of them are fake," she lamented. "We urge you to listen to official channels."

The videos depicted a range of fabricated scenarios, from fake news bulletins to images of severe flooding, roaming sharks, and poignant scenes of human suffering.

Others seemed to show locals—often with marked Jamaican accents, seemingly intended to reinforce stereotypes—partying, sailing, jet-skiing, or joyfully swimming before the hurricane's arrival.

Such false content overwhelms important safety alerts and leads users to underestimate the danger.

Fake, yet convincing

"Fake content undermines the serious work of the government," asserts Amy McGovern, a meteorology professor at the University of Oklahoma, specializing in the use of AI for extreme weather forecasts. "They will lead to loss of life and property."

The clips identified by AFP are mainly circulated on TikTok. And while the platform is supposed to mandate that AI-generated content be labeled as such, only some of them bore the obligatory tagging.

TikTok appeared to have removed more than two dozen of these videos, along with several accounts dedicated to sharing them, after AFP's report. However, other examples were found on Facebook and Instagram, where the rule is supposed to be the same.

For Hany Farid, co-founder of cybersecurity firm GetReal Security and a professor at the University of California, Berkeley, Hurricane Melissa illustrates how new text-video models have indeed "accelerated the spread of convincing fake content."

New applications allow for the generation of hyper-realistic human avatars. Many users seemed unaware that the images in question were fake despite their watermark, as observed by AFP in the comments on the videos.

Like this old man shouting he wouldn't "move for a little breeze" in a TikTok AI video. "God, please protect grandfather's house and the mango tree," wrote one commenter below, while another requested more updates on the state of his property.

A flood of similar prayers appeared under another video depicting a woman calling for help, holding babies under a roofless house.

"The paradox of the information age is that we become less informed as a public while the amount of information increases," pointed out Hany Farid to AFP.

AFP