Google was forced to urgently deactivate the AI image generator based on the Nano Banana model, integrated into Google Earth. The tool lasted less than a day — such was the swift community reaction to the frankly dangerous content that the neural network created without any restrictions.
The functionality allowed users to "draw in" any objects or events on the map based on a text description. However, within the first hours of operation, generated images depicting the aftermath of an explosion in Los Angeles and a flooded U.S. Capitol leaked online. These are not just harmless artifacts — such fakes could trigger panic or be used in disinformation campaigns.
Particularly alarming is the fact that the system did not filter even the most sensitive queries. Open data researcher Henk van Ess publicly confirmed: his prompts, including generating images of refugees at the Mexican border or nuclear facilities in Iran, went through without a single warning or block. This is a direct violation of Google's own policies against harmful content.
The developers explained the removal of the tool by stating that users were massively violating the service's rules. However, in my view, the problem runs deeper: the company clearly underestimated the risks associated with generative models in a geospatial context. When a neural network gains access to real coordinates, the line between fiction and "documentary" evidence blurs catastrophically fast.
This incident is a vivid example of how the hasty deployment of AI without proper oversight can turn into a reputational blow. Google should have first tested moderation mechanisms on a closed sample of users before releasing such a powerful tool to the public. Otherwise, we will see such failures again and again — and not only from giants like Google.