Google shuts down AI deepfake tool in Google Earth after one day

Google Earth’s experimental AI image editing feature—powered by the Nano Banana 2 model—was live for less than 24 hours before Google shut it down following widespread misuse. The tool allowed users to generate AI-altered satellite images from text prompts, which were watermarked as AI-generated but still managed to bypass content moderation. Dutch researcher Henk van Ess demonstrated the tool’s vulnerabilities by creating misleading images, including fake depictions of refugees near the Mexican border and bomb craters in Gaza, which were not flagged by Google’s initial safeguards.

Google initially defended the feature, citing its usefulness for geospatial professionals, but reversed course after users shared AI-generated screenshots that violated its policies. The company now admits the feature’s rapid rollout lacked adequate guardrails, despite watermarking generated images. Van Ess also highlighted how the AI-generated content could evade detection tools like Hive AI, raising concerns about misinformation spread through seemingly credible sources like Google Earth.