WASHINGTON, United States (AFP) — Artificial intelligence (AI) generated videos were clogging social media feeds Monday as Hurricane Melissa surged toward Jamaica, diverting attention from critical safety information about the massive Category 5 storm.

AFP surfaced dozens of fakes — most bearing watermarks for OpenAI’s text-to-video model Sora — as Melissa was set to pummel the Caribbean island with violent winds and heavy rains.

The videos depicted a range of fabricated scenarios, from dramatic newscasts and shots of severe flooding to images of sharks in the water as well as poignant scenes of human suffering.

Others appeared to show locals — often voiced with strong Jamaican accents that seemed aimed at reinforcing stereotypes — partying, boating, jet skiing, swimming or otherwise minimizing the threat of what forecasters have warned could be the island’s most violent weather on record.

Senator Dana Morris Dixon, Jamaica’s information minister, said that she and other ministers were jointly taking part in a Monday press briefing to give “correct information” about the approaching monster storm.

“I am in so many WhatsApp groups, and I see all of these videos coming. Many of them are fake,” Dixon said. “And so we urge you to please listen to the official channels.”

Even ostensibly innocuous fakes can contribute to drowning out important safety alerts or cause viewers to underestimate the danger of severe storm events, experts said.

“This storm is a huge storm that will likely cause catastrophic damage, and fake content undermines the seriousness of the message from the government to be prepared,” said Amy McGovern, a University of Oklahoma meteorology professor whose research has focused on using AI to improve extreme weather forecasting.

“Eventually such fake content will lead to loss of life and property,” she told AFP.

The clips AFP identified spread primarily on TikTok, where only some carried a label despite the platform’s policy requiring users to disclose realistic AI-generated content.

TikTok appeared to remove more than two dozen such videos — as well as multiple accounts dedicated to sharing them — after AFP flagged them to the platform.

A few other examples were found circulating on Facebook and Instagram, even though parent company Meta’s policies say labels are also mandated for photorealistic videos created with AI.

Hany Farid, co-founder of the cybersecurity company GetReal Security and a professor at the University of California, Berkeley, said the hurricane-related AI content underscores how new text-to-video models have “accelerated the spread of convincing fakes”.

The apps allow users to create clips featuring hyper-realistic human likenesses. AFP reached out to OpenAI for comment, but did not get an immediate response.

Many viewers seemed unaware the images were AI-generated, despite the Sora watermark, AFP’s review of the videos’ comment sections found.

“God please protect grandpa’s home and mango tree,” one commenter wrote under an AI video on TikTok of an old man yelling at the hurricane that he would not “move for a little breeze”.

Another user asked him to post more updates on the state of his property.

A rush of similar prayers were offered under a different video that portrayed a woman crying for help while holding babies under a roofless home.

“The paradox of the information age is that we are becoming less informed as a public as the amount of information increases,” Farid told AFP.