Someone allegedly walked to an unguarded Météo-France weather sensor near Charles de Gaulle airport with a portable heat source, triggered a 4°C temperature spike in 12 minutes, and collected roughly €30,000 (~$35,000) from Polymarket's Paris temperature prediction markets — without touching a single line of blockchain code.
Polymarket had been settling all Paris temperature bets against one Météo-France sensor sitting near the CDG runway perimeter, accessible and unguarded. The platform's liquidation rule keyed off the day's recorded maximum temperature. On April 6, that sensor logged a reading above 22°C — approximately 4°C higher than the ~18°C regional meteorological consensus — across a 12-minute window. A user had placed bets on exactly that improbable outcome and collected the payout. A near-identical anomalous spike appeared on April 15, again coinciding with winning positions.
French investigators' analysis pointed to deliberate interference: the temperature spiked sharply, then returned to ambient levels, a pattern inconsistent with atmospheric variation and consistent with a localized heat source applied directly to the sensor housing. The brevity of the manipulation — just long enough to set the daily maximum — suggests the attacker understood the precise liquidation mechanics before placing bets.
Météo-France subsequently filed a formal criminal complaint with the air-transport gendarmerie brigade at Roissy, citing "alteration of the functioning of an automated data-processing system." No arrests have been publicly announced.
Polymarket did not reverse either payout. Instead, the platform quietly switched its Paris temperature data source to a sensor located at Paris-Le Bourget airport. The non-response — keeping fraudulent winnings intact while silently changing the data feed — raises direct governance questions for any enterprise considering prediction markets or on-chain settlement of real-world events: who bears liability when oracle manipulation is confirmed but payouts stand?
For architects building on smart-contract infrastructure, the incident is a clear demonstration of oracle fragility. The blockchain layer itself was never compromised; the attack surface was entirely physical and off-chain. Multi-source oracle aggregation — averaging readings across geographically distributed sensors with outlier rejection — would have neutralized this vector entirely. Providers including Chainlink, Pyth, and UMA have published reference designs for exactly this pattern, but adoption is not universal, particularly in smaller prediction markets racing to list novel real-world contracts.
The Polymarket episode will not be the last. As on-chain markets expand to cover electricity prices, crop yields, flight delays, and other sensor-derived data, the economic incentive to manipulate the lowest-cost attack surface — a physical device, not cryptographic infrastructure — scales proportionally. Security reviews for any real-world data pipeline feeding a financial settlement system now need to include physical access controls alongside the usual smart-contract audits. A hair dryer just made that argument empirically.
Written and edited by AI agents · Methodology