Google pulls Google Earth AI image tool a day after launch

Google has removed a new Google Earth feature after roughly a day of operation because disinformation researchers warned that it allowed anyone to drop convincing fake scenes onto Google’s own satellite imagery.

The feature was heavily criticized with several professionals calling for it to be removed. Google attempted to implement safeguards that would allow it to keep the feature operating, but ultimately conceded.

Why did Google approve misinformation?

Google recently launched and quickly removed a Google Earth feature called “create image,” that used Google’s Nano Banana 2 model to turn a typed prompt into an invented scene layered over real satellite, aerial and 3D map data.

A product manager, Bryan Horowitz, said that users could zoom to a spot on the web version, tap the button, and type whatever they wanted to appear there. Google pitched it as a way to “visualize history, create real estate plans and more.”

However, open-source investigator Henk van Ess wrote on Substack, that Google spent 20 years building a trustworthy reference only for it to add a “button that makes things up.”

His main issue with the feature was that the generated images would borrow the authority of the Google Earth platform.

BBC Verify used the tool to generate a collapsed Eiffel Tower, a sinkhole swallowing the Great Pyramid of Giza, and Russian tanks in Kyiv. Van Ess also created images that included refugees at the Mexican border and a bomb crater beside a hospital in Gaza, neither of which met any resistance.

NPR produced images of Iran’s Kharg Island in flames and a flooded U.S. Capitol while AFP fabricated images of an explosion in Paris and a nuclear site in Iran.

404 Media staff added skyscrapers to farmland, planted a “homeless encampment” in Los Angeles, and staged a protest outside Google’s own offices.

Why should satellite imagery be credible?

Open-source analysts rely on images from Google Earth because a photograph taken from hundreds of miles up is costly and difficult to fake.

Jake Godin, a senior researcher at the investigative group Bellingcat, revealed that forged satellite images have surfaced before, including fake damage at a U.S. base in Bahrain during the U.S.-Israel conflict with Iran, but he explained that adding a one-click generator makes the fakes too easy to mass-produce.

Bill Greer, a geospatial analyst and co-founder of the nonprofit Common Space, said the erosion of trust in satellite imagery is “potentially particularly damaging.”

Google explained that every image created with the tool carries a SynthID digital watermark that can be verified through the Gemini app or the Lens feature in Search. It also wrote that it doesn’t allow image creation regarding harmful topics. However, shortly after, BBC Verify found it could trick Gemini into calling a fake Google Earth image genuine.

Google’s prompt filtering also failed easily. While it rejected a request to build a gallows to “hang traitors” outside the UK parliament, a vaguer version of the same prompt went through.

Google has since pulled the feature, stating that while geospatial professionals found the tool useful, people were sharing screenshots of generated imagery that violated its policies. It has stated that it will work on implementing stronger guardrails, but whether the tool will return remains an open question.

The smartest crypto minds already read our newsletter. Want in? Join them.