Google yanks AI satellite image generator from Google Earth after 24 hours and outcry
The tool let users type any prompt to fabricate satellite views. Researchers quickly created fakes of bomb craters, refugee camps, and nuclear plants, prompting a rollback and promises of stronger guardrails.
Google on 31 July withdrew a Google Earth feature that had allowed users to generate synthetic satellite imagery with text prompts, one day after it was announced. The rapid reversal followed a wave of demonstrations by open-source researchers and journalists who showed how easily the tool could produce realistic-looking fakes.
The launch and immediate backlash
Google product manager Bryan Horowitz announced the integration of the Nano Banana 2 image generator into Google Earth on 30 July, inviting users to "type whatever you want to see." Within hours, open-source investigator Henk van Ess had generated and shared fabricated scenes: a non-existent nuclear plant in Iran, refugees at the US-Mexico border, a bomb crater beside a Gaza hospital, and a car crash on an Amsterdam street.
The forgery does not have to look convincing on its own. It inherits the credibility of the map it was born on.
What the tool could generate
The same day, multiple news outlets replicated the exercise. NPR produced images of Iran's Kharg Island on fire and a flooded US Capitol. BBC Verify showed a collapsed Eiffel Tower, a sinkhole swallowing the Great Pyramid of Giza, and Russian tanks in Kyiv. El Periódico generated a blast crater in Melilla and an invasion scene at Plaça de Catalunya in Barcelona, complete with small fires and collapsed buildings. 404 Media added a fictitious explosion crater in Los Angeles and placed protestors outside Google's Mountain View campus.
Watermarks and detection failures
Google initially defended the tool by stating every generated image carried a SynthID digital watermark and that harmful topics were blocked. The company suggested users verify images through the Gemini app or Lens. Tests by van Ess and BBC Verify undermined that defence: van Ess fooled an AI-detection tool (Hive) with a video generated in Google Earth, while the BBC showed it was possible to trick Gemini into declaring a fake Google Earth image real.
Journalists might be able to check that image and report on it. But again, not everyone is going to do that.
Expert concerns about satellite imagery trust
Satellite imagery has long been a trusted reference for verifying events in conflict zones, natural disasters, and human rights investigations. Researchers such as Jake Godin of Bellingcat called it "a safe bet" because it was hard to fake. The new tool, by embedding generation directly into the mapping platform, changed that calculus. Sam Stockwell of the Alan Turing Institute said the tool "corrupts the very measure used to debunk fakes." Brady Africk of the American Enterprise Institute noted that the fakes were "anchored to real satellite images, removing some of the tell-tale signs of AI generation like mismatched terrain and scale."
- Google announces integration of Nano Banana 2 into Google Earth.
- Henk van Ess creates and posts fabricated satellite images on X.
- Company says images are watermarked with SynthID and harmful topics blocked.
- Google withdraws the feature, citing policy violations and pledging stronger guardrails.
Google's response and rollback
On 31 July, Google reversed course. A company statement acknowledged that "people uniquely trust Google Earth for a reliable view of the world" and that users had been sharing generated imagery that appeared to violate its policies. The feature was rolled back while stronger guardrails are built. The statement added that the generated images did not appear in the main Google Earth experience for others to see and were watermarked, but that was not enough to prevent immediate policy violations. No timeline was given for a possible return.
