Google Pulls Its AI Image-Editing Tool from Google Earth After Users Faked War Zones and Border Scenes
A feature that let anyone repaint satellite photos with text prompts lasted less than 48 hours before Google yanked it. The reason: researchers showed it could generate convincing fake imagery of real places, and at least one AI detection tool was fooled.
Key points
- Google launched an AI image-editing feature inside Google Earth on a Thursday, then disabled it the following day after abuse reports surfaced.
- The tool used a model called Nano Banana 2 to alter real satellite photos based on typed instructions, such as adding a bomb crater near a hospital or refugees at a border crossing.
- Researcher Henk van Ess of Digital Digging tested the tool and found no prompts were blocked or softened during his session.
- Van Ess also showed that a video made with the feature fooled Hive, a commercial AI-content detection service, despite Google's watermark being present.
- Google says generated images were never visible to other Google Earth users, but screenshots shared on social media did not carry that context.
Google Earth has always felt like a window onto the real world. On Thursday, Google opened that window a little too wide.
The company quietly launched a feature letting users type a description and watch AI repaint any patch of satellite imagery. Want to see what a solar farm looks like on that empty field? Fine in theory. Want to drop a fake bomb crater next to a Gaza hospital, or add a crowd of refugees at the Mexican border? Also, it turned out, fine in practice.
That second category is the problem.
What exactly did the tool do?
It let anyone with Google Earth type a text prompt and get back an altered version of real satellite imagery. The underlying model was called Nano Banana 2. Google told users the images carried a digital watermark confirming they were AI-generated, and said the system blocked "image creation on harmful topics."
Henk van Ess, a researcher at Digital Digging who tracks how AI tools can be misused, ran his own tests and told The Verge AI those guardrails simply did not hold. "Nothing was refused, nothing was softened, and nothing suggested I try a different prompt," he wrote, sharing examples including fabricated humanitarian crisis scenes placed over real map coordinates.
He then went one step further. Van Ess took a video produced inside the tool and ran it through Hive, a paid service that businesses use to detect AI-generated content. Hive missed it. The watermark, in other words, did not make the image unfakeable, and the detection industry had not caught up.
Why did Google shut it down so fast?
Speed matters here. Within 24 hours, Google issued a second statement admitting the problem and pulling the feature. The company said: "We know that people uniquely trust Google Earth for a reliable view of the world... we've also seen people sharing screenshots of generated imagery that appear to violate our policies. So we're rolling back this feature while we work on implementing stronger guardrails."
A crucial detail buried in that statement: the fake images never appeared inside the main Google Earth view that other users see. They existed only in a personal session. But screenshots stripped that context away the moment anyone posted them online.
What does this mean for ordinary people?
For most users, nothing changes day to day. Google Earth looks the same. The risk is less about your own experience and more about photos you might see shared on social media claiming to show a real place.
If a satellite image appears somewhere surprising, showing damage or crowds at a specific location, it is worth pausing before sharing it. Reverse image search tools can help check whether an image matches known, verified sources.
Google says it will bring the feature back once it has built "stronger guardrails." What those look like remains to be seen.
Common questions
Can I still use Google Earth normally?
Yes, completely. Google pulled only the experimental AI image-editing feature. The main Google Earth app and website, including standard satellite and street-level views, are unaffected.
Does a watermark mean an AI image is safe to trust?
Not on its own. Van Ess's test showed a watermarked video from the tool still fooled a commercial AI-detection service. Watermarks are a useful signal, but they are not a guarantee, and they disappear the moment someone screenshots and crops an image.
