Google Earth AI tool pulled one day after launch over fake imagery fears

Google removed a Google Earth AI tool on July 31, one day after launching it, after journalists and open-source researchers showed it could generate realistic-looking satellite images of events that had not happened. The quick reversal shows how trust and safety problems can slow even high-profile AI features when the product is tied to a service people use as a view of the real world.

The feature launched July 30 on the web version of Google Earth. It let users pick a location, select “create image,” and use a text prompt to generate a scene with Google’s Nano Banana model, according to Google and reports from 404 Media and NPR.

In a statement posted on X, Google said people place unusual trust in Google Earth as a dependable picture of the world. The company said some geospatial professionals had found helpful uses for the feature, but other users were sharing generated screenshots that appeared to break Google’s policies. Google said it was rolling back the feature while it adds stronger guardrails.

Why did Google remove the Google Earth AI tool?

The concern was that the tool made fake satellite-style evidence easy to create and share. A deepfake is synthetic media that can make a false event or person appear real; in this case, critics warned that fabricated overhead imagery could be mistaken for documentation of wars, disasters, protests, or attacks.

404 Media reported that the tool could create a blast crater in Los Angeles and place protesters outside Google’s Mountain View campus. NPR generated images showing Iran’s Kharg Island on fire and the U.S. Capitol flooded, scenes that would be major news if genuine. Open-source researcher Henk van Ess told NPR that he tested prompts involving refugees at the Mexican border and a nuclear plant in Iran, and that the tool did not reject them.

Researchers raised a broader problem: satellite images are often used to verify breaking news because they have historically been hard to fake convincingly. Bellingcat researcher Jake Godin warned that one-click generation could make false images easier to produce and faster to spread. He also said governments could use the existence of such tools to cast doubt on real images.

What did Google say about safeguards?

Google had pointed to SynthID, its watermarking system for AI-generated content, as one protection. The company said images made with the feature carried that marker, which can identify them as AI-generated in tools such as Gemini, and said it blocks harmful topics.

Critics were not reassured by that explanation, according to the reports. Their concern was practical: screenshots can move across social platforms detached from context, and many viewers will not run a watermark check before reacting to an image.

Google said it will bring the feature back only after adding stronger protections. The company did not give a timeline for restoring it.

This story draws on original reporting from Decrypt.