New AI Ranks Fake Disaster News by How Much Harm It Could Cause

When a hurricane or wildfire strikes, social media fills with rumors, bad advice, and outright lies. Current tools can flag whether a post is false, but they can't tell you if a claim is just silly or genuinely life-threatening. This new research goes a step further by measuring the severity of each false claim.

The system works in two stages. First, it automatically pulls out false claims from posts that mix text, images, videos, and links. Then, it checks each claim against evidence and rates it on two scales: believability (will people actually trust this?) and harmfulness (what happens if they do?). The researchers tested it using Reddit posts about real hurricanes and wildfires, with human reviewers helping set the scoring standards.

The most interesting finding: large language models—the same kind of AI behind ChatGPT—were far better than traditional software at agreeing with human judgments about how dangerous a claim is. The approach that worked best was giving the AI examples of past claims and the decisions humans made on them, so it could follow the same reasoning rather than just guess a score.

Why does this matter? During a disaster, seconds count. If an AI can automatically flag the most harmful misinformation—like a fake shelter location or a bogus evacuation route—emergency agencies can correct it fast and prevent panic. It's still early, and the system was tested only on a limited set of events, but it points toward a future where AI helps communities stay safe from both the storm and the lies around it.