Just a day after its high-profile debut, Google was forced to shut down the AI image generator integrated into Google Earth. The tool, built on the Nano Banana model, allowed users to "fill in" any objects or events on the map based on a text prompt—from fantastical landscapes to politically sensitive scenes. However, the rapid spread of content violating the service's rules forced the corporation to urgently disable the feature.
The problem turned out to be systemic. Within the first hours of operation, generated images appeared online depicting an explosion crater in Los Angeles and a flooded U.S. Capitol. These fakes instantly spread across social media, creating the illusion of real events. Moreover, as independent open-data researcher Henk van Ess discovered, prompt moderation was virtually nonexistent.
In his tests, van Ess requested the generation of images related to sensitive topics—such as scenes with refugees at the Mexican border or nuclear facilities in Iran. None of these requests were blocked, casting doubt on the effectiveness of the claimed filtering algorithms. This is not just a technical glitch but a signal of deep gaps in the ethical protocols of AI development.
Google's decision to remove the feature so quickly is a rare case of admitting its own mistake. The corporation explained that the tool "did not meet safety standards" but did not disclose details. Clearly, the risk of reputational damage and potential legal consequences outweighed the benefits of innovative marketing.
This incident is a stark reminder that generative models working with geodata require a fundamentally different level of control. Unlike text-based chatbots, here an error is visualized in a form that many perceive as documentary evidence. For the crypto industry, where trust in data and transparency are cornerstones, this case is especially instructive: decentralized content verification systems could be a solution, but for now, the market watches as even giants stumble on flat ground.
My conclusion: the hasty launch and weak moderation are signs that the race to implement AI is outpacing the development of accountability mechanisms. Until such tools undergo independent audits and receive built-in verification "watermarks," they cannot be trusted in the context of real events.