Google Earth's New AI Image Tool Is Already Fooling Fake-Photo Detectors

Key points

• Google launched Nano Banana 2, an AI image-generation feature inside Google Earth, globally on the web version earlier this week.
• Researcher Henk van Ess of Digital Digging used it to create fake satellite images showing refugees near the US-Mexico border and a bomb crater beside a hospital in Gaza.
• Google says every image the tool creates carries a SynthID digital watermark, an invisible tag embedded in the image to identify it as AI-generated.
• Digital Digging was able to fool Hive, a commercial AI-detection service, using an AI-altered video produced with the same Google Earth feature.

Google Earth has a new trick. Type a text prompt, basically a written instruction, and the platform's Nano Banana 2 feature will generate a fake satellite image layered on top of real aerial and 3D map data. Google pitched it in a blog post as a way to visualise historical sites or preview building projects. Within days of launch, a researcher had already used it to make fake crisis images.

What did the researcher actually make?

Henk van Ess, an open-source intelligence researcher at Digital Digging, created two images that look like genuine satellite footage: one showing what appeared to be refugees gathered near the Mexican border, and another showing a bomb crater next to a hospital in Gaza. Neither scene was real. Both looked plausible at a glance.

That is the risk here. Some outputs are obviously fake, like a Sphinx-Statue of Liberty hybrid. But the tool can produce images convincing enough to share on social media before anyone checks.

Does Google's watermark fix the problem?

Not entirely. Google told The Verge AI that every Nano Banana 2 image includes a SynthID watermark, the same invisible signature Google uses across other AI image tools. You can check a suspicious image by asking the Gemini app (Google's AI assistant) or using Google Lens in Search.

The trouble is that watermarks can be missed or worked around. Digital Digging tested an AI-altered Google Earth video against Hive, a well-known AI content detection service, and Hive did not flag it as artificial.

What should you do if you see a satellite image online?

Treat satellite imagery in news stories or social posts the same way you would treat an unverified eyewitness photo. Van Ess recommends cross-checking any suspicious image against independent satellite platforms such as Sentinel-2 or Landsat, both of which are publicly accessible. He also suggests checking the orbital metadata, the technical record of when an image was taken and by which satellite, to see whether the numbers add up.

For a quick first check, paste the image into Google Lens or tag it with "@verifyai" inside the Gemini app.

Satellite imagery has long carried unusual authority. A photo of a city street can be staged; a satellite view feels like proof from space. Nano Banana 2 changes that assumption, and the misinformation risk does not wait for detection tools to catch up.