Google handed users the easiest possible tool for fake satellite imagery, then pulled it after two days
Google integrated its Nano Banana 2 image model into Google Earth as a way for users to generate imagery directly within a familiar mapping interface. The feature was short-lived: within two days of its release, Google pulled it after users publicly demonstrated how simple it was to create fake but plausible satellite photographs using nothing more than a text prompt.
The examples that surfaced were pointed. In one widely circulated case, a user directed the model to fill an empty lot near the Mexican border with a column of refugees. The result was a convincing overhead image that, without context, could be mistaken for genuine satellite data. The concern is not merely aesthetic - fabricated geospatial imagery has clear potential for misuse in disinformation campaigns, where a doctored overhead shot can be used to falsely document troop movements, border activity, environmental damage, or infrastructure changes.
Satellite imagery carries a particular kind of authority. Audiences tend to treat it as objective, physical evidence in a way they might not treat a standard photograph. That perception makes it a more potent vehicle for misinformation, and it raises the stakes for any tool that makes such imagery easy to produce. Traditional manipulation of satellite photos requires significant technical skill; a prompt-based tool removes that barrier almost entirely.
Google has not issued a detailed public explanation of its decision to pull the feature, but the speed of the withdrawal suggests the company recognized the oversight quickly. The episode is a pointed example of the gap that can open up between the capabilities of a generative model and the safeguards needed before deploying it in a context with obvious potential for harm. As AI image generation becomes more tightly woven into everyday software - including tools people associate with factual, real-world data - the question of where and how these models are deployed becomes as important as the quality of the output itself.
