Google Earth has been a trusted window into our planet for two decades. Journalists, investigators, and open-source intelligence (OSINT) researchers rely on it daily to verify events, monitor conflict zones, and debunk falsehoods. So when Google announced on July 30 that it was integrating its AI image generator Nano Banana 2 directly into Google Earth, the reaction was swift — and furious.
The idea sounded harmless enough. Type a prompt, and the tool would generate custom images using Google’s real satellite, aerial, and 3D imagery as a foundation. Want to see what your neighborhood looked like a century ago? Visualize a real estate project? Bring ancient Pompeii to life? That was the pitch.
But within hours, researchers showed exactly how this could go wrong.
What actually happened
OSINT researcher Henk van Ess was among the first to sound the alarm. With a single sentence, he generated a fake nuclear power plant in Iran, a refugee camp on the US-Mexico border, and a hospital in Gaza with a bomb crater beside it. His question summed up the disbelief: “What on earth is Google doing?”
Others followed suit. BBC Verify created images of a collapsed Eiffel Tower, a sinkhole swallowing the Great Pyramid, and Russian tanks in Kyiv. NPR generated a flooded US Capitol and fires at an Iranian oil terminal. Users fabricated aerial strikes, war zones, and natural disasters with nothing more than a text prompt.
The tool went live on a Thursday. By Friday, Google had rolled it back. Less than 24 hours.
Why the backlash was so intense
Google Earth isn't just another app. It's a foundational tool for verification. When a satellite image comes from Google Earth, it carries instant credibility. As van Ess put it: “The forgery does not have to look convincing on its own. It inherits the credibility of the map it was born on.”
That’s the core problem. Attaching AI-generated fiction to genuine coordinates, on genuine imagery, blurs the line between fact and fabrication in a way that independent AI detection tools couldn't always catch. Even Google’s own SynthID watermarking and Gemini verification could be bypassed.
For OSINT researchers who depend on unaltered satellite photography, the feature was a nightmare. Bellingcat founder Eliot Higgins responded sarcastically: “No way this could be abused.”
The deeper lesson
Google said it was pausing the feature to implement “stronger guardrails”. The company acknowledged: “We know that people uniquely trust Google Earth for a reliable view of the world.”
But the episode raises an uncomfortable question: Should generative AI be embedded in tools that people rely on as evidence? Placing a photorealistic image generator inside a service trusted as a record of the physical world — and expecting only harmless fun — shows a remarkable naivete.
Google hasn’t permanently killed the feature. It’s paused. When and if it returns, the safeguards will need to be far more robust. But the trust erosion may already be done.
For now, one thing is clear: in the age of generative AI, even the sky isn’t safe from deepfakes.
Tags:
Technology