It took less than a day for Google to kill its own feature. On Thursday, the company launched a tool letting users alter Google Earth satellite imagery with simple text prompts, powered by an AI model called Nano Banana 2. By Friday, it was dead.
The problem was painfully predictable. Researcher Henk van Ess immediately demonstrated how the tool could be weaponized, generating images that added refugee camps near the Mexican border and a bomb crater beside a hospital in Gaza. Google’s initial defense—that images were watermarked and the company blocked harmful topics—crumbled quickly. Van Ess showed he could fool Hive’s AI detection tool with generated video, and noted that in his testing, “Nothing was refused, nothing was softened, and nothing suggested I try a different prompt.”
Google’s reversal came with a candid mea culpa. “We know that people uniquely trust Google Earth for a reliable view of the world,” the company stated, acknowledging that while geospatial professionals found useful applications, bad actors were already sharing policy-violating screenshots. The company pulled the feature entirely, promising “stronger guardrails” before any relaunch.
The episode highlights a recurring industry blind spot: the gap between what a tool can do and what people will actually do with it. Watermarking, the standard safety fig leaf, proved useless in practice. The generated images never appeared in the main Google Earth experience, but screenshots don’t need to—they’re shareable in seconds. For a platform built on cartographic trust, even a day of deepfake geography is a day too long.
