Google Earth's new AI image generation function didn't…

Google suspended a new AI image-editing feature for Google Earth less than 24 hours after its launch when users began generating realistic, deceptive imagery. The tool, which utilized the Gemini Nano Banana 2 model, allowed users to transform real-world locations using text prompts.

What happened

The feature launched on a Thursday, enabling users to modify satellite views with generative AI. Almost immediately, the system was used to create convincing deepfakes of sensitive and catastrophic scenarios. AI researcher Henk van Ess demonstrated the tool's capabilities by generating images of fake refugee camps along the US-Mexico border and a nuclear plant in Iran. Other users leveraged the function to produce imagery depicting nonexistent floods and a 9/11-style attack. One specific instance involved the generation of a kaiju attacking Downtown Los Angeles.

By Friday, Google reversed course and pulled the feature. A spokesperson stated that while geospatial professionals used the function for legitimate purposes, others created imagery that violated company policies. The company emphasized that the AI-generated images were watermarked and never appeared in the public Google Earth experience, but noted that the feature was removed to allow for the implementation of stronger guardrails.