Google pulled the Google Earth AI tool a day after launching it, after users showed how easily it could be used to manufacture convincing-looking satellite views of real places. The feature let people use text prompts with Google’s Nano Banana 2 model to alter Google Earth imagery, creating synthetic geospatial scenes that could be screenshotted and shared outside the product.
Google said in a public statement Friday that it was rolling back the feature while it adds stronger safeguards. The company said people “uniquely trust Google Earth for a reliable view of the world,” and said it had seen screenshots of generated imagery that appeared to violate its policies.
The company also said the AI-created images did not appear in the main Google Earth experience for other users and were labeled with an AI-generated watermark. That distinction matters inside Google’s product. It matters less once a screenshot or video leaves the interface, which is where the usual internet garbage disposal unit takes over.
Why did Google remove the Google Earth AI tool?
Google removed the feature after examples circulated showing that prompts could generate misleading scenes tied to sensitive real-world locations and events. Henk van Ess of Digital Digging said he intentionally tested the system by creating altered imagery that added refugees near the Mexican border and a bomb crater near a hospital in Gaza.
On Thursday, Google responded to van Ess by saying images made with Nano Banana 2 in Google Earth carried a digital watermark and that the company blocked “image creation on harmful topics.” By Friday, Google’s position had changed: the tool was being rolled back because some shared generations appeared to break its rules.
Van Ess said his testing did not run into meaningful resistance from the prompt system. He wrote that nothing was refused, softened, or redirected to another prompt. He also said he created a video in Google Earth that fooled Hive’s AI detection tool, an example of the gap between a watermark policy and what happens when generated media is repackaged.
How the feature worked
The mechanism was straightforward: a user could ask the AI model to modify a view in Google Earth, and the system would render an altered version of that location. Because the base material was satellite-style imagery from a product people associate with maps and documentation, the output had a different risk profile than a random fantasy image generator.
Google said some geospatial professionals had used the feature for useful purposes, though it did not spell out those uses in the statement. The company’s rollback does not mean the generated images were inserted into Google Earth for everyone. It means Google decided that letting individual users create and share altered Earth imagery needed more guardrails than it shipped with.
The episode is a familiar AI product failure mode: launch a prompt box, add policy language, assume labels will carry more weight than virality, then discover that screenshots ignore product boundaries. Google has not said when, or whether, the Google Earth image-generation feature will return.
This story draws on original reporting from The Verge.
