The feature, which used Google’s Nano Banana 2 image generation model, was rolled out on Thursday and disabled nearly 48 hours later, after the company became aware of screenshots of generated images that appeared to depict locations altered in ways that violated Google’s policies. While the company does not specify what prompted its intervention, it notes users “have a strong expectation of Google Earth as a source of authoritative information about the world.”
Following the removal of the feature, BBC Verify was able to recreate several examples of altered scenes using the tool, including the Eiffel Tower lying in ruins, a sinkhole engulfing the Great Pyramids of Egypt and Russian tanks poised to enter Kyiv. AI and misinformation expert Henk van Ess was also able to demonstrate the ability to create misleading images of real world locations, including a fake nuclear power plant in Iran, a refugee camp along the US and Mexico border and a hospital in Gaza with a crater.
“Not only do the images have questionable value as evidence, but the very act of associating them with real-world locations and Google’s own satellite imagery adds an element of credibility to the deception,” said Van Ess.
Google stated that images generated by its AI model contain invisible watermarks and directed users to Gemini and Google Lens to analyze images and detect authenticity. However, BBC Verify was able to uncover ways to bypass these measures, as well as manipulate the prompt to avoid detection. Meanwhile, researchers found some AI detection tools were unable to identify images generated by the Google Earth tool.
Henry Ajder, an AI detection researcher, noted that images of populated places and battlespaces could cause “incredible damage to populations if they were to appear as credible evidence of events on the ground.”
“The danger comes when the situation on the ground is unclear or time-sensitive, and people are looking for reliable information,” he added.
Geospatial analyst Bill Greer added that imagery of the Earth has long been considered a “trusted source” of information by both governments and the public, meaning its misuse could undermine confidence in the technology and its ability to provide truthful insight.
The episode underlines the challenge facing both creators and users of AI imagery, as the ability to generate increasingly realistic images threatens to erode the value of other trustworthy sources of information.
