Google Earth, now with a side of potentially harmful fantasy

TL;DR: A new AI image generation feature now lives inside Google Earth. An independent researcher found that it’s trivially easy to abuse, letting you create realistic satellite images of violent or politically inflammatory scenes, despite the company saying it has guardrails.

What happened: With a simple prompt, you can now generate AI images that drop fake structures, people, or events over Google’s satellite imagery. The new feature went live yesterday in Google Earth using Nano Banana 2—and it’s a puzzling add-on to a go-to tool often used in education and by open-source intelligence analysts and journalists.

While making realistic-looking deepfakes of satellite images was possible before, this AI feature does it in one step (often in just a minute)—and results can be refined to look more convincing. You can probably see the potential danger from space, and journalists and researchers are already voicing their concerns.

What could go wrong?: Google’s own pitch for the tool is benign—recreating historical places or visualizing a real estate project using the actual topography before you break ground. Its blog post encourages you to “let your imagination run wild,” but that’s exactly the worry.

Researcher Henk van Ess found that a single sentence was enough to place a nuclear power plant in Iran, drop a crowd of refugees near the US-Mexico border, or even depict a fatal crash on an Amsterdam street. 404 Media reported it was able to add both a homeless encampment and a bomb blast crater in Los Angeles.

Google’s response: In a statement to van Ess, Google said it takes misinformation seriously and that “every image created with Nano Banana in Google Earth includes the SynthID digital watermark.” But detecting that watermark, which is invisible to the human eye, takes an extra step—either asking Gemini or using Google Lens to see if the image is fake. (There’s also a Gemini watermark on every image, but it’s easy to crop out.) And when van Ess posted a screen recording of his fake imagery on X, a popular third-party AI detector gave it just a 0.8% likelihood of being AI.

Google also said it prevents “image creation on harmful topics,” but van Ess reports that none of the prompts he tried were refused. As of publication time, Tech Brew could still prompt the tool to place a nuclear plant in Iran without any objection.

Bottom line: Google says there are guardrails in place. But given everything, it’s unclear what it counts as a harmful topic. In an age of so much disinformation, having an AI image generator inside this reputable digital atlas sounds like a recipe for disaster. —WK

Also at Google…

• Finally! One of Chrome’s most annoying update features could be gone soon.