For One Day, Google Made It Easy to Fake Satellite Images

For about 24 hours starting on a Thursday, Google Earth gave its billions of users a powerful new tool powered by artificial intelligence. The feature let anyone type a few words and create a fake but very realistic satellite image of any place on Earth. Within one day, Google turned the feature off after people used it to make dangerous fake images, raising fears about the spread of false information online.

Google said it wanted the tool to spark people's imaginations. For example, a user could type a prompt to see what their neighborhood looked like 100 years ago, or what an empty lot could be turned into. The company even showed off fun examples, like making the Google campus in California look like a scene from a science-fiction movie.

But things quickly went in a different direction. Researchers, journalists, and other users discovered there were no safeguards to stop harmful or misleading images from being made. A researcher named Henk van Ess created fake images of a nuclear plant in Iran, a bombed hospital in Gaza, and a car crash in Amsterdam — all of which looked completely real. National Public Radio also tested the tool and generated images showing a flooded U.S. Capitol and fires at an oil terminal in Iran.

What made this so alarming was how easy it had become. Before this tool existed, creating a fake satellite image took a long time and required special skills. Now, anyone could do it with a single sentence. The fake images were especially convincing because they matched the lighting, colors, and camera angles of real Google Earth photos. 'A forgery built that way doesn't have to be convincing on its own,' van Ess wrote, meaning the trust people already have in Google Earth does the work for the fake.

Google said that every AI-generated image included a hidden digital watermark — a kind of invisible label identifying it as fake. However, users found that taking a simple screenshot could sometimes fool AI detection tools into missing that label. Google called sharing those screenshots a violation of its rules. On Friday, the company announced it was disabling the feature while it worked on adding better safety protections, hinting that the tool could come back in the future.

This situation reminded many people of a serious real-world example from 2014. After Malaysia Airlines Flight 17 was shot down over Ukraine, killing nearly 300 people, Russia's government showed satellite images at a news conference claiming it was not responsible. Investigators at the online research group Bellingcat used real Google Earth images to prove that Russia's photos were fakes. They showed the images were doctored and that a Russian missile had actually fired the shot. That investigation showed how much people trust Google Earth — and why putting that trust at risk is so dangerous.