AI
Google's AI Image Tool in Earth Raises Misinformation Risks Despite Safeguards
The search giant's new Nano Banana feature can generate convincing fake satellite imagery, and researchers have already demonstrated how it could spread false claims about real-world events.
The Verge
Google has rolled out an AI image-generation tool embedded directly into Google Earth that lets users create synthetic satellite and aerial views simply by typing a description. The feature, called Nano Banana, launched globally on the web this week—but security researchers are already highlighting troubling implications for how fabricated imagery could fuel misinformation campaigns.
What the tool does#
The new capability transforms text prompts into photorealistic flyover videos and images that appear to be sourced from Google Earth's satellite, aerial, and 3D imagery libraries. Google markets it for benign uses: previewing historical sites or visualizing proposed real estate developments. The company applied its SynthID digital watermark to all generated content and emphasized these guardrails when challenged over potential misuse.
The misinformation problem#
Researchers from Digital Digging, including Henk van Ess, have already created synthetic images depicting false scenarios—refugees clustered near the Mexican border and a bomb crater adjacent to a hospital in Gaza—to demonstrate how readily the tool can manufacture convincing false evidence of events that never occurred. While some obviously absurd outputs (like a Sphinx merged with the Statue of Liberty) are easy to dismiss, many images appear sufficiently authentic to mislead casual observers.
The watermark and detection mechanisms, though present, offer imperfect protection. Van Ess managed to bypass at least one AI detector by processing a Nano Banana video, suggesting that determined actors could obscure the tool's traces.
Verification strategies#
Google recommended using its "@verifyai" feature within Gemini to flag suspect images, or turning to Google Lens in Search for AI-detection checks. Van Ess proposed additional safeguards: cross-referencing questionable imagery against independent satellite platforms like Sentinel-2 or Landsat, and validating metadata such as capture dates and satellite source information.
