Google Earth’s new AI generator already faked Gaza bomb craters and border refugees

A text prompt is now all it takes to warp reality inside Google Earth. The platform’s new generative AI feature, powered by what appears to be a model called Nano Banana 2, lets users create synthetic satellite and aerial flyover images on demand. Researcher Henk van Ess of Digital Digging wasted no time stress-testing the tool, generating chillingly realistic scenes of “refugees near the Mexican border” and a bomb crater beside a hospital in Gaza. The results are not cartoonish deepfakes. They look like intelligence briefing slides.

Google launched the feature globally on the web version of Earth earlier this week, pitching it as a creative tool for visualizing historical sites and real estate developments. Some outputs are easy to spot as fabrications—a Sphinx crossed with the Statue of Liberty is hardly fooling anyone. But the problem sits squarely in the uncanny valley of plausibility. Van Ess proved you can manufacture evidence of humanitarian crises in minutes, using a platform that carries the implicit trust of satellite imagery. That trust is now a liability.

Google’s defensive posture relies on SynthID, its invisible digital watermark, and a verification workflow using the Gemini app or Lens. “We take misinformation seriously,” a Google spokesperson said, pointing to the watermark. Yet van Ess already showed the guardrails have cracks. A video he altered with Google Earth’s AI slipped right past Hive’s AI detector. Relying on users to manually verify every suspicious aerial shot with orbital data from Sentinel-2 or Landsat—as van Ess himself recommends—is a technical fix for what is fundamentally a social and psychological problem.

The speed of the abuse matters here. The tool went live this week, and within days, it produced synthetic images of active conflict zones. That timeline should alarm anyone who tracks how quickly generative tools are weaponized. Watermarks are a sensible layer, but they are not a solution when the images are being ripped from context and shared on platforms where nobody checks metadata. Google gave everyone a blueprint for building fake satellite intelligence. Now we get to see how fast it scales.