From a 'Collapsed' Eiffel Tower to Fake Nuclear Sites: Google Earth's New AI Feature Frozen Within 48 Hours
Google has suspended its new AI feature in Google Earth less than 48 hours after its launch, due to concerns that it could create fake satellite images for spreading disinformation. The feature integrated the Nano Banana 2 image generator and allowed adding objects onto real satellite imagery. BBC tests created fictional scenarios, including a collapsed Eiffel Tower. Experts warn about the high credibility of the fakes.
Google has suspended its new artificial intelligence feature in Google Earth less than 48 hours after its launch, following concerns that it could enable the creation of fake satellite images that might be used to spread disinformation. The feature, which integrated the Nano Banana 2 image generator, allowed users to use text prompts to add objects, structures, and events that never occurred onto real satellite imagery, aerial photos, and 3D views. The images could then be downloaded and shared on social media. In a statement released yesterday (Friday), Google said it is pausing the feature "while we work on implementing stronger safeguards," after identifying users sharing images created with it in ways that could violate company policy. "We know that people place unique trust in Google Earth as a source that presents a reliable image of the world," the company said. In tests conducted by the BBC, researchers were able to use the tool to create a series of fictional scenarios that looked particularly credible, including a collapsed Eiffel Tower in Paris, a massive sinkhole swallowing the Great Pyramid of Giza, and Russian tanks in central Kyiv. Experts warned that the danger lies not only in the quality of the images, but in the fact that they are "welded" onto a real location on Google Maps. According to them, the combination of a real satellite image with AI-generated content gives the fake particularly high credibility. Disinformation expert Henk van Ess, who examined the feature, also managed to create a fictitious nuclear power plant in Iran, a non-existent refugee camp on the US-Mexico border, and a mock hospital in the Gaza Strip with a crater next to it that looked like the result of a bombing. "The fake doesn't have to be convincing on its own. It inherits the credibility of the map on which it is created," he said. Google noted that every image created with its AI tool includes an invisible watermark, and that their origin can be checked using Gemini or Google Lens. However, according to the BBC, in some cases it was possible to bypass the detection mechanisms and even cause Gemini to identify fake images as real. Some external detection tools also failed to identify that the images were AI-generated. Additional tests showed that some of Google's safeguards could also be bypassed by slightly changing the wording of the prompts. For example, a request to create a "traitor-hanging facility" near the British Parliament was blocked, but a more general wording was accepted and the image was created. A request to add the word "Trump" to the White House lawn was rejected, but the tool agreed to create a community garden in the exact same spot. According to Henry Ajder, a researcher specializing in identifying AI-generated content, the ability to produce fake satellite images of war zones or disasters could be particularly dangerous. "When events unfold quickly and information is limited, such images could destabilize and mislead both journalists and the public," he said. Bill Greer, a geospatial analyst and co-founder of the non-profit satellite organization Common Space, also warned about the erosion of public trust. According to him, satellite imagery has been considered a particularly reliable source of information for years, but the proliferation of fakes alongside reduced access to real satellite data by governments could erode that trust. "The more fake images we see and the fewer real ones, the harder it will be to know what to trust," he said.