AI or real? BBC analyses viral China disaster videos

Stephen McDonell, the BBC's man in China, has been picking apart a wave of disaster clips flooding Chinese social media, and what he found isn't reassuring. As extreme weather has battered parts of the country this year, so has a parallel flood of synthetic video showing storms, floods, and collapsing buildings that never actually happened. Some of it is obviously fake if you look closely. A lot of it isn't.

That gap matters because these clips don't stay online as curiosities. They get reposted, stripped of context, and treated as breaking news by people who have no way of checking. McDonell notes that Chinese authorities have already promised a crackdown on this kind of misinformation, which tells you how seriously officials are taking the risk of panic, wasted emergency resources, or plain public confusion during actual weather crises.

The harder problem, and the one McDonell keeps circling back to, is detection. AI video generation has gotten good enough that the old tells, warped hands, flickering shadows, physics that doesn't quite add up, are showing up less often and are easier to miss when they do appear. Even trained eyes, including the BBC's own team, are having to slow down and analyse frame by frame rather than trust a quick glance.

China isn't unique here, it's just an early and vivid example. Any country with widespread social media use and a steady diet of real extreme weather is a candidate for the same problem: a public primed to believe disaster footage because disasters really are happening more often, and a technology that can now manufacture convincing footage of disasters that aren't.