AI in Google Earth Makes It Possible to Fabricate Fake Satellite Images

On Thursday, Google added an AI-powered feature to the web version of Google Earth that lets any user paint onto satellite images things that don't exist in reality. Just point the cursor at the spot you want and type in a text prompt — and a made-up object appears on top of Google's real imagery. The feature runs on the Nano Banana model.

According to the outlet 404 Media, the tool's capabilities open the door wide to disinformation. Google Earth used to be a valuable tool for OSINT analysts tracking changes on the ground during conflicts and disasters. Now that same service can be turned into a machine for producing fakes.

What It Managed to “Draw”

During their tests, the 404 Media staff added skyscrapers to a rural area, a “homeless camp” to a part of Los Angeles where homelessness has become a hot-button political issue, an “explosion crater” to another part of the city, and even generated protesters outside Google's own corporate buildings.

OSINT researcher Henk van Ess described his own experience like this: with a single sentence he placed refugees at the Mexican border, “built” a nuclear power plant in Iran, and staged a fatal accident on a street in Amsterdam — all of it on top of Google's real satellite imagery. “What on earth is Google doing?” he wonders.

Another specialist, Benjamin Strick, notes that if you enter your office address and ask it to “make it look like a drone strike happened here,” the system produces an image very much like real footage from Ukraine, Russia or the Middle East. In his view, this will make verification work dramatically harder.

Google's Response

In response to an inquiry, the company pointed to its post on social media. Google claims it takes disinformation seriously: every generated image is tagged with a SynthID digital watermark, and an image's origin can be checked through the Gemini app or the Lens feature in search. On top of that, the company blocks generation on “harmful” topics and promises to improve its safeguards.

However, as 404 Media points out, these measures ignore the obvious: most people won't bother verifying whether an image is genuine, and those who spread fakes aren't the least bit troubled by the prospect of being caught — the damage is done first. Strick adds that even watermarks can be bypassed: “This is bad.”

Source: 404 Media