Google disabled its new AI image generation tools feature in Google Earth on the web less than 24 hours after launch, following demonstrations that it could create misleading content, including a fictitious nuclear plant in Iran. The rollback, announced on July 31, 2026, came after investigative journalist Henk van Ess used the tool to generate imagery of refugees near the Mexican border and that fake nuclear facility. The feature had allowed users to reimagine Google Earth imagery based on text prompts powered by Google's Gemini AI model.
A Feature Built for Education, Not Deception
Google had pitched the tool as a way to visualize history and data. Intended uses included showing Pompeii before the eruption of Vesuvius or turning landmarks into illustrated infographics with data pulled from Gemini. The company framed it as an educational enhancement to a platform that millions trust for accurate satellite views of the planet.
But the same capability proved dangerously flexible. Van Ess, a journalist known for open-source investigations, showed how easily the tool could fabricate plausible scenes. His demonstrations, reported by NPR and Futurism, highlighted a core problem. Creative or misleading interpretations can become especially convincing when grounded in Google's real-world data. A fake nuclear plant in Iran, rendered in satellite style, looks like evidence.
Google's Own Words on the Rollback
Google announced the rollback in an update to its original announcement on The Keyword, its official blog. The company acknowledged the trust users place in the platform. The update stated: "We know that people uniquely trust Google Earth for a reliable view of the world. We've seen geospatial professionals using this feature for a range of useful purposes, however we've also seen people sharing screenshots of generated imagery that appear to violate our policies. So we're rolling back this feature in Google Earth while we work on implementing stronger guardrails. It's important to note that generated images didn't appear in the main Google Earth experience for others to see and were watermarked as AI generated."
That last point, the watermark, proved weaker than Google implied. The company used SynthID, its AI watermarking system, to invisibly mark generated images. But the system was easily defeated. Taking a photo of a generated image stripped the watermark's reliability. Hive AI, an independent detection service, rated a screen-recording of Google Earth's AI-generated images as only 1% likely to be AI-generated video. That number underscores how fragile the protection was.
Watermarks Are Not Enough
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
The SynthID failure matters beyond this single feature. Watermarking is a common method for identifying AI content, but it can be circumvented with simple tricks. A screenshot, a photo of a screen, or a screen recording can erase or obscure the invisible marks. In this case, the 1% rating from Hive AI showed that even a basic recording defeated detection.
There is a world of difference in impact between educational infographics and potentially believable images of fictitious geopolitical events. Google's own examples, like Pompeii or infographics, are benign. But the same engine can produce a fake refugee crisis or a phantom nuclear site. Coupling real-world satellite data with photorealistic image generation creates unprecedented misinformation risks. That combination is what made the tool so compelling and so dangerous.
What Remains Unclear
The exact improvements to safety guardrails are not clear. Google has not detailed what stronger measures it will implement. As of publication, the AI image generation feature in Google Earth remained paused. Testing Google Earth on the web revealed that previously generated AI images remain accessible, even though new generation is disabled.
Google will have to navigate carefully if the feature is to remain both useful and safe. The company has been integrating generative AI across its products, from search to mobile. This rollback shows the limits of that push when applied to geospatial data. SynthID, as it stands, is not enough to prevent simple manipulations of Google's AI images from being passed off as real.
The timeline is tight. On July 30, 2026, Forbes reported on the feature's addition. By July 31, 2026, Google had pulled it. That speed suggests the company recognized the severity of the problem. But the pause leaves questions open. Will the feature return with better filters? Can any filter stop a determined user from misusing a tool that generates photorealistic satellite imagery on demand?
For now, Google Earth remains a reliable view of the world, minus the AI reimagining. The episode serves as a reminder that generative AI, when fused with trusted data sources, demands more than good intentions. It demands guardrails that actually hold. Google's next move will determine whether this tool becomes a safe educational asset or stays on the shelf.