Google disabled a new AI image-editing feature in Google Earth less than a day after its release. the tool, powered by the Nano Banana 2 model, was removed after users generated deceptive imagery of conflict zones and sensitive borders.
The 24-hour lifespan of Nano Banana 2 in Google Earth
Google integrated the Nano Banana 2 image-generation model into Google Earth to allow users to overlay fabricated scenarios onto satellite, aerial , or 3D imagery. According to the report, the company intended for the software to help students visualize history, assist in professional real estate planning, and allow users to visualize projects. However, the feature was pulled on Friday, within 24 hours of launch, after Google observed users sharing screenshots that violated company policies.
The rapid reversal highlights a significant gap between Google's intended use cases and the actual behavior of its user base. While Google noted that these generated images were watermarked and did not appear in the main Google Earth experience for other users, the ability to create and download these images provided a turnkey kit for creating convincing misinformation.
From Gaza bomb craters to Iranian nuclear plants
The vulnerability of the Nano Banana 2 model was exposed by Hank van Ess, an OSINT expert and author of the Digital Digging newsletter. As the report says, van Ess successfully used the tool to "plant" a nuclear plant in Iran and place refugees near the Mexican border. Most critically, he demonstrated that the AI could generate a fabricated bomb crater at a hospital in Gaza without any intervention from the system.
Van Ess noted that despite Google's claims of preventing image creation on harmful topics, the software provided no warnings, no softened output, and no refusals for these highly sensitive prompts. this failure suggests that the initial filters for the Nano Banana 2 model were insufficient to handle geopolitical flashpoints, allowing users to create visual "evidence" of war crimes or military expansions with a few keystrokes.
The shadow of fake drone strikes in Qatar and Burj Khalifa fires
This incident occurs amid a broader trend of AI-generated geospatial disinformation. In March, the social media platform X saw the spread of AI-generated satellite images that falsely depicted a successful Iranian drone strike on a US radar system in Qatar. Similarly, a fabricated video showing Dubai's Burj Khalifa engulfed in flames reached tens of millions of views, illustrating how easily the public can be misled by AI-generated imagery of landmarks.
The risk is exacerbated by the financial incentives of social media. The report mentions that X has had to warn creators that they may face temporary suspension from its monetization scheme if they post AI-generated videos of armed conflict without proper labels. Because the program rewards high engagement, there is a built-in incentive for bad actors to use tools like Nano Banana 2 to create shocking, fake imagery that drives views and profit.
What 'stronegr guardrails' will actually stop Hank van Ess?
Google stated it is rolling back the feature while it works on "implementing stronger guardrails," but it remains unclear what these technical barriers will look like.. A primary question is whether Google can possibly blacklist every sensitive coordinate on Earth—ranging from hospitals in Gaza to borders in Mexico—without rendering the tool useless for its legitimate educational purposes.
Furthermore, it is unknown if the watermarks tied to the Gemini AI model are robust enough to survive basic editing or cropping, which is common when images are shared on platforms like X. For a service like Google Earth, which is 21 years old and trusted by millions for a reliable view of the world, the challenge is no longer just about image quality, but about preventing the platform from becoming a factory for believable lies.
Comments 0