Google disabled an artificial intelligence image generator built into Google Earth less than 48 hours after announcing it, after the tool was used to create "deepfake" aerial images, the American tech giant confirmed.
The feature, powered by Google's Nano Banana 2 image generator, had let users alter satellite views of any location on the planet. Google shut it down on Friday, July 31, one day after it was rolled out.
Google Earth has let people explore the planet through a three-dimensional virtual globe for 20 years, combining satellite imagery, aerial photographs and topographic data. Open-source intelligence researchers and journalists around the world use it daily to verify facts and debunk false information.
How the AI tool worked
Users could zoom in on any location, click "Create image" and type a description of what they wanted to see, altering the satellite view of that spot. The AI tool then generated an image by blending genuine satellite pictures and aerial photographs with the requested content.
At launch, Google had encouraged users to experiment by imagining what their neighbourhood looked like a century ago or might look like a century from now, by generating an AI depiction of the ancient city of Pompeii, or by "reimagining" famous landmarks.
Instead, AI researchers quickly demonstrated how easily the tool could be misused to spread false information while appearing to be legitimate content produced by Google itself.
Safeguards found lacking
Google had said built-in rules in its AI model would block the creation of images depicting "harmful subjects." But AI specialists soon found no protective mechanism actually stopped problematic content from being generated.
Google also said every AI-generated image carried a digital watermark identifying it as synthetic. But journalists at BBC Verify, the British public broadcaster's disinformation unit, found that a simple screenshot was enough to strip some images of the watermark and evade AI-detection tools.
Fake nuclear plants and submerged landmarks
Henk van Ess, an AI researcher and disinformation expert, exposed the risks by generating images of a nonexistent nuclear power plant in Iran, a refugee camp on the US-Mexico border, and a fake hospital in Gaza with a bomb crater nearby.
National Public Radio separately produced a fabricated image showing the US Capitol submerged in water in Washington DC, as well as a fictional fire at a strategic oil terminal in Iran. BBC Verify's own tests generated images of Russian tanks in the Ukrainian capital, Kyiv.
Fears over conflict zones
The creation of fake images in conflict zones is a particular concern, especially when events are moving quickly and verified information is scarce. An expert interviewed by the BBC said journalists might be able to check such an image and report on it, but not everyone would take that step.
Google's response
In a press statement, Google said it had seen geospatial professionals using the feature for genuinely useful purposes, but had also noticed people sharing screenshots of generated images that appeared to breach its rules.
The company said it would work on putting protective measures in place, suggesting the feature could eventually return.
