Google pulls Google Earth AI image tool after 24 hours over misinformation

Google pulls Google Earth AI image tool after 24 hours over misinformation

Google has disabled a new artificial intelligence tool for Google Earth just 24 hours after its global launch on Thursday. The feature allowed users to create visual stories by editing real satellite imagery using an image generator, but it was quickly withdrawn after users created deceptive content involving war zones and geopolitical crises.

The tool, powered by Google's Nano Banana AI model, was intended for benign uses. Google suggested the feature could allow a teacher to visualize a site as it appeared in 78 AD or enable a user to transform a car park into a garden. Users accessed the tool via a Create Image tab on the web version of the application, where they could input prompts to edit specific sites.

However, critics and researchers quickly demonstrated the tool's potential for misinformation. AI researcher Henk van Ess reported using a single sentence to place refugees near the Mexican border, plant a nuclear plant in Iran, and simulate a fatal crash on a street in Amsterdam. In the Amsterdam example, van Ess prompted the AI to insert a traffic accident and a crowd in high-resolution, resulting in an image of a red car smashed into the side of ROBO-BAR in De Weteringschans with glass covering the road.

Other tests revealed further vulnerabilities. The news outlet 404 Media found the tool could add a bomb blast and crater to Los Angeles in approximately 10 seconds. The BBC produced images of Russian tanks in Kyiv, Ukraine, a sinkhole swallowing the Great Pyramid of Giza, and a collapsed Eiffel Tower.

Van Ess warned that the danger lies in the perceived authenticity of the imagery. He noted that while such fakes were possible with Photoshop 20 years ago, the barrier was once skill and effort, but is now negligible.

"A satellite image isn’t a memory, it’s a record – courts, insurers and war crimes investigators treat it as evidence of what stood on a patch of ground on a given day,"

Henk van Ess, AI researcher, via Metro

Van Ess argued that because the forgeries are welded to real coordinates and match the lighting and angle of genuine captures, they inherit the credibility of the map itself.

Google responded on Friday, stating that it takes misinformation seriously. The company noted that all AI-generated content includes SynthID, an invisible watermark that the Gemini chatbot can detect. In a statement on X, Google explained that while geospatial professionals used the feature for useful purposes, some users shared screenshots that appeared to violate company policies.

Google emphasized that these generated images did not appear in the main Google Earth experience for other users to see. The company is now rolling back the feature to work on stronger guardrails.

This incident follows a similar rapid reversal by Meta last week. Meta was forced into a 24-hour U-turn after launching an AI tool that could edit public Instagram posts, a move the company later admitted missed the mark after users were required to manually opt out to prevent their photos from being remixed.

Van Ess suggested that AI should remain on the reading side of satellite imagery rather than the painting side, questioning if safeguards can remain effective as images degrade through repeated screenshotting.