AI or real? BBC analyses viral China disaster videos

BBC's verification team spent time picking apart a batch of disaster clips circulating online that supposedly show catastrophic events in China. The goal was simple on paper: figure out which ones are genuine footage and which ones were cooked up by generative AI tools. In practice, that job has gotten a lot messier over the past year or so.

The reason is obvious once you think about it. Video generators like Sora and Veo have gotten good enough that motion, lighting, and physics no longer give away the fakes the way they used to. A flooded street, a collapsing building, a fire tearing through a market — these are exactly the kind of chaotic, high-emotion scenes that spread fast on social platforms, and they're also exactly the kind of scenes AI models have gotten better at rendering convincingly.

BBC's approach leaned on the usual toolkit: checking metadata, cross-referencing weather and geography, looking for reused footage from unrelated events, and examining small inconsistencies that AI models still tend to produce, like warped text on signage or physically impossible reflections. Some of the clips checked out as real disasters, repurposed with false claims about location or date. Others showed clear signs of synthetic origin once examined closely.

What makes this particular case notable isn't the specific videos themselves but what it says about the state of viral misinformation generally. A year ago, most people could spot an AI video with a bit of attention. That gap is closing fast, and outlets like the BBC are now treating video verification as a full investigative discipline rather than a quick sanity check.