A rapid launch followed by an equally swift shutdown — that is how one could describe the story of the AI image generator based on the Nano Banana model, which Google integrated into Google Earth. The tool lasted less than 24 hours before developers were forced to deactivate it. The reason is mundane yet telling: users instantly found ways to violate the service's rules, turning the mapping product into a platform for creating blatantly misleading content.
The essence of the feature was to "draw in" any location or event directly on the map based on a text prompt. It sounded like a convenient creative tool for visualizing hypothetical scenarios. However, reality proved far less harmless. Within just a few hours of the launch, AI-generated images spread across social media, depicting, for example, a giant explosion crater in Los Angeles or a flooded U.S. Capitol.
What is particularly alarming is that the system failed even basic query moderation. Open data researcher Henk van Ess discovered during his tests that his prompts, including requests to create images of refugees at the Mexican border or a nuclear facility in Iran, went through without any blocks. This is direct evidence that content filtering mechanisms in Nano Banana were either absent or configured extremely superficially.
From a professional analysis standpoint, this incident is not merely a technical glitch but a symptom of a systemic problem. Google is once again trying to integrate generative AI into mass-market products, but the rush leads to reputational risks. When it comes to mapping services, which are perceived as a source of objective data, any manipulation of the visualization of reality is especially dangerous. Fake images of disasters or military installations could trigger panic or be used for propaganda.
My verdict: Google did the right thing by rolling back the feature, but the very fact of such a launch raises questions about internal quality control processes. In the pursuit of innovation, one must not forget basic responsibility to users. The lesson for the entire industry is obvious: generative AI in geospatial products requires not just technical implementation but also a deep development of ethical and legal constraints. Otherwise, we risk getting a tool that does not inform but manipulates.