Google has withdrawn a newly launched Google Earth feature that let users alter satellite imagery with AI text prompts, only a day after introducing it. The decision matters because Google Earth is widely trusted as a representation of real places, while the tool could create convincing but false depictions of sensitive events and locations.

Tests by researcher Henk van Ess generated scenes including refugees near the Mexican border and a bomb crater beside a Gaza hospital. Google said outputs were watermarked and not shown in the main public Google Earth experience, but acknowledged screenshots appeared to breach its policies; it is pausing the feature while developing stronger safeguards. Van Ess also said a generated video bypassed Hive’s AI detector and that his prompts were not refused or moderated.