As generative AI progresses, fake images of tornadoes are becoming increasingly convincing and more common. While the digital dupes may seem harmless, Western University-based Northern Tornadoes Project warns that they can cause undo confusion, fear, and danger.
The program’s director, David Sills, says he first started seeing the AI images in 2025. “Someone had taken a photograph of a storm and inserted a tornado into it,” he says. “We knew it was AI because we did manage to find the original photo, so we could see exactly what was changed.”
Sills says they had been expecting something like this to start happening as AI improved, but that this year there’s been a flood of fake images shared on social media that spiked in the wake of tornado warnings issued by Environment and Climate Change Canada (ECCC).
In June, while an area near London, Ont., was under a tornado watch, there were four different AI photos circulating that showed a storm that was much more severe than what actually took place. “A couple of them were actually photos of certain intersections or certain roadways, certain neighborhoods, and someone inserted a tornado into it,” says Sills.
Cottage Q&A: If we don’t have a basement, where should we go during a tornado?
Sometimes the photos are so exaggerated that it’s easy for Sills to identify them as a fake, but the trickier ones can be time consuming to investigate. “It can take as long as two hours,” he says. “We have to kind of reconstruct the photo: what direction it’s facing, where it was taken, and then look at how that relates to radar, and what we were seeing there and in other storm photos at the time, to figure out that there was no way there was a tornado there.”
Sills says it’s a growing problem that will likely get worse; he’s already seeing the first AI-generated tornado videos during storms in Ontario. The danger, he says, is that it can muddle ECCC forecasters’ ability to accurately issue tornado warnings.
“When they’re thinking they may issue a tornado warning, they’re waiting for some evidence from a report of a funnel cloud or something on radar. And then suddenly this image pops up in the social media feeds of a tornado in the location they were considering. At that moment, they have to decide whether it’s real or not,” says Sills. “And it’s really difficult to do that. Especially within 10 minutes, which is the target lead time for a tornado warning.”
Forget the movies: meet a real storm chaser
The concerns are two-fold: if the ECCC posts a tornado warning based on an AI photo, people may take action that isn’t necessary. And if the photo is real, but forecasters believe it to be AI-generated and decide not to issue a warning, this could put people in harm’s way.
Sills says this likely isn’t the intent of the people creating these AI images. “I think there’s certain people doing it for attention or clicks, and they don’t realize what they’re doing is problematic to the extent that it’s illegal,” he says. “In other cases, it seems to be someone just sharing it with a friend as a joke, and then that friend passes it to someone else, and it quickly loses its context and suddenly is out there.”
In one case, a media outlet in London picked up the photo and published it thinking it was real, says Sills. “Once a media organization is tricked and they put it out, it’s really hard to contain.”
Related Story What is Environment Canada’s new “polygon” severe weather alert system?
Related Story A new provincial partnership aims to increase emergency preparedness in cottage country
Related Story The cottager’s guide to cloud-collecting