As Storm Boelo abated in the North Yorkshire coastal town of Whitby, publican Sue Taylor went out for a walk. Near the town’s famous abbey she found a collapsed stone wall, with rocks from it sitting neatly on the bonnets of adjacent cars. She took two photographs and posted them on Facebook when she got back to her office, without any editing.
Then dozens of comments told her the images had been created or altered using AI.
What actually happened
Taylor, 48, who runs the Black Horse Pub, saw the damage on her normal dog-walking route. The permit-only parking spaces were used by people living nearby or at the side of the landmark 199 Steps route up to the abbey.
“It was a shocker,” she said. “I used to live up near the abbey for about 12 and a half years, and although we experienced some strong winds, I’d never seen it be strong enough for a wall to be blown over during the night.”
On the accusation that she had faked it: “It’s not something that I would ever have considered altering, let alone know where to start.” She described the online doubt from internet sleuths as “a bit ridiculous”.
The BBC has viewed independently-posted photographs and videos of the same damage online, and spoken to eyewitnesses who corroborated Taylor’s images. English Heritage, which operates Whitby Abbey, commented on the incident. One of the damaged cars was still in place on Monday afternoon.
Why people got it wrong
Prof Theocharis Kyriacou, an expert on AI at York St John University, said he could understand some viewers’ scepticism, but urged caution — particularly as “real people” had been affected.
“Some of the rocks appear to have fallen into rather neat or unusual arrangements, which might seem surprising,” he said. “However, we should remember that most of us have probably never witnessed this particular type of structural collapse before. Something looking unusual or unexpected does not necessarily make it artificial.”
That is the sharpest line in the story, and it generalises. People are not actually judging whether an image is generated. They are judging whether it matches their expectation of how the world normally looks — and almost nobody has a mental reference library for what a wall blown over in the night does to a row of parked cars.
Kyriacou’s test is better than intuition: photographs from different sources, independent footage and eyewitness accounts are “more meaningful than individual impressions of what looks real or fake”.
The cost, named
He said the Whitby photographs illustrated a wider issue, as trust in visual evidence was being eroded.
“For generations, photographs have been regarded as relatively reliable representations of reality,” he said. “Now, because of the misuse of generative AI, even genuine photographs are increasingly being questioned. That is a rather sad consequence of what is otherwise a remarkable technological achievement.”
This is the half of the problem that gets less attention. Most coverage of generated imagery is about fakes being believed. The Whitby case is the opposite failure, and in some ways the more useful one to plan for: once everybody knows images can be fabricated, a true photograph loses its power to settle anything. Anyone with a reason to deny a real event now has a free, universally understood excuse.
Both failures in one county
North Yorkshire has had the other kind too. Earlier this year a fake video appeared to show Pete Kilbane, City of York Council’s Labour deputy leader, handing cash to men in balaclavas, with an unspecified flag appearing behind him. He was given police advice to step up security at his home.
Kilbane said he was “shocked” when he first saw it: “This is an attempt to undermine democracy and make legitimate, factual debate impossible.”
Last month, University of York students raised concerns that planned changes to the licence agreement for the Turnitin plagiarism system could see their work used to train AI models. The university said it would review any changes; a Turnitin spokesperson said its policy would not be updated any earlier than September 2027, to “give us time to discuss any changes with customers”.
Healthy scepticism, and the other thing
Kyriacou pointed to AI-generated images and video posted during deadly floods in Nepal as a “particularly troubling example” of misuse, and encouraged people to “question what they encounter online rather than accept everything at face value”.
“But there is an important distinction between healthy scepticism and automatically dismissing something as fake simply because it looks extraordinary.”
For practical purposes: detection tools help only at the margins. Google’s public AI-detection portal reliably identifies only images watermarked by participating generators, which a camera photograph of a real wall obviously is not. Corroboration — a second photographer, a video, a named witness — remains the only test that works, and it is the one Taylor’s photographs passed.
Related, from the other direction: a generated photograph that did convince a records office.
