Weather data sabotage: the climate attack surface enters the radar

In plain terms

Weather data - ground sensors, radiosondes, radars, satellites - now feed much more than TV bulletins. Energy trading, flight paths, grid, precision agriculture, and increasingly, AI forecasting models (GraphCast, Pangu-Weather, WITT urban, #1044). MIT Technology Review documents as of July 17, 2026, the rise of the risk of sabotage of this chain - physical and software.

Context

The weather observation infrastructure is a decades-old public-private patchwork. Ground stations are poorly defended, often without strong authentication of flows, and exchange protocols (WMO Information System, GTS) assume implicit trust. Yet the same data now serves: regulators (grid), traders (spot energy, ag), insurers (parametric), and AI models trained on open histories.

What's changing

The risk is no longer "the station failure" but selective injection: a few biased sensors that shift a storm forecast, a grid signal, a parametric insurance trigger. In the AI era, this risk is escalating - a model learned on poisoned data propagates the error far beyond the station.

What's concrete today

• Physical: isolated sensors, without remote monitoring, where local intrusion suffices.
• Chain: little or no cryptographic signatures on inter-agency flows.
• AI: the "AI nowcasting" layer consumes the same raw data as a numerical model, without systematic anomaly detection at input.

Scenarios

• Most likely: isolated incidents, difficult to distinguish from a sensor failure. The real issue is regaining confidence afterward.
• Hardest: coordinated multi-site sabotage around an extreme event (hurricane, blackout) - sufficient offset to bias an operational decision.
• Most useful for a sophisticated attacker: data-supply-chain attack - poisoning the training history of weather AI models.

Safeguards

Signatures of flows, cross-redundancy (satellite/ground/private), anomaly detection on the weather operator side, mandatory audits for AI models that feed critical systems. None of this is in place at scale.

So what

Weather sabotage is a useful reminder: the AI data supply chain starts before the dataset. Touching the probe means touching everything that depends on it. To be integrated into data-supply-chain risk mappings - on par with dataset scandals and model backdoors.