Historia
agosto 6, 2026
Google Earth’s AI Editor Was Pulled After Deepfake Fears Exploded
Google’s Nano Banana 2 editor promised creative ways to reimagine places, but researchers quickly showed it could fabricate convincing crises on trusted satellite imagery. The company rolled it back within a day, saying stronger safeguards are needed.
Google Earth’s latest AI experiment collided with the one thing its users expect from the platform: a credible view of the real world. Within a day, Google withdrew a prompt-based image editor after examples showed how easily it could turn trusted satellite imagery into persuasive fiction.
The feature, powered by Nano Banana 2, was introduced as a creative layer for Google Earth on the web—one that could visualize a city’s past or imagine a new community basketball court.
1 Google also promoted uses including real-estate concepts and reconstructions of ancient Pompeii.
But researchers immediately tested the darker version of that promise. Independent investigator Henk van Ess said a single prompt let him place “refugees near the Mexican border,” then add a nuclear plant in Iran and a fatal crash in Amsterdam, all over Google’s own imagery.2 The concern was not merely that AI can make fakes—many tools already can—but that the fakes were being built directly atop a platform journalists and open-source investigators use to establish what is real.
Google initially pointed to SynthID watermarks and said it blocked harmful requests. Critics argued those protections were too brittle for the way misinformation actually travels: as screenshots, compressed clips and reposted images. Van Ess said his testing found, “Nothing was refused, nothing was softened, and nothing suggested I try a different prompt.”3
By the next day, Google had rolled the tool back. The company acknowledged that professionals had found legitimate uses, but said screenshots of generated scenes “appear to violate our policies” and that it would build “stronger guardrails” before proceeding.3 Google stressed that the altered images did not enter the main Earth experience for other users and were marked as AI-generated.
The swift reversal leaves both sides with a point: generative imagery can support visualization and education, but placing it inside a widely trusted reference tool risks giving fabricated events the authority of a map.