China’s Ministry of Emergency Management announced a breakthrough AI system last week designed to automatically detect fake disaster footage before it spreads online. The system works by analyzing video metadata, pixel patterns, and user engagement metrics. It has a 94% accuracy rate, according to the press release, which nobody independently verified.
What the system actually does is flag videos that go viral. All of them. Real typhoon footage gets the same treatment as deepfaked mudslides and that one cat that fell off a roof and became a meme in 2019. The algorithm cannot distinguish between a genuine flood and a TikTok trend because both involve water and movement and people screaming.
Why build an AI system that solves nothing when you could just hire people to watch videos? Because people cost money every month and AI costs money once, plus venture capital, plus government contracts, plus a podcast about the future of disaster prevention.
The real problem—that misinformation spreads faster than corrections—remains unsolved. The fake solution has already been deployed to seventeen provinces. It is currently flagging weather reports as suspicious because they contain the word “unprecedented,” which appears frequently in both real emergencies and YouTube thumbnails.
The BBC managed to distinguish real from fake by watching them. The AI cost 8 million yuan and requires constant retraining. The BBC used journalists who have seen rain before.