Google retire en urgence l'IA de Google Earth après la…

On July 30, 2026, Google introduced a new AI-powered feature to the web version of Google Earth, allowing users to generate ultra-realistic visual elements directly over real satellite, aerial, and 3D data. Intended for use by educators and geospatial professionals, the tool was removed within 24 hours following widespread concerns regarding its potential for misinformation.

What happened

The feature utilized a model called Nano Banana 2, which allowed users to select a specific geographic coordinate and describe a scene for the AI to render. Because the generated images were anchored to real-world coordinates and utilized authentic geographic data, the results were exceptionally credible.

Researchers and users quickly demonstrated the tool's capacity for creating geopolitical and dramatic deepfakes. Demonstrations included the generation of fake refugee camps at the Mexican border, non-existent nuclear facilities in Iran, fabricated bomb craters near hospitals in Gaza, and artificial disaster scenes in Amsterdam and Washington, D.C.

While Google had implemented its SynthID digital watermarking technology to identify AI-generated content, the measure proved insufficient. Users frequently shared screenshots of the imagery without verifying their authenticity, and in several instances, Google's own internal detection tools failed to identify the images as synthetic. Google stated that while these images did not appear in the standard Google Earth experience for other users, the rapid circulation of screenshots necessitated the tool's emergency withdrawal. The company is currently reviewing its safety guardrails before considering a potential future release.