Google announced on Friday it was rolling back a newly launched AI image-generation tool for Google Earth after widespread criticism over its potential to spread disinformation. The feature, which allowed users to create AI-generated images based on satellite imagery, was removed less than 24 hours after its debut.
Google confirmed the reversal in a statement on X, citing concerns that users had exploited the tool to generate misleading or policy-violating content. The company stated it was temporarily disabling the feature while developing stronger safeguards. A Google spokesperson said the tool had been used for both legitimate purposes and policy violations, including the creation of deepfaked disaster scenes and fabricated geopolitical imagery.
The AI tool, powered by Google’s Nano Banana 2 image-generation model, integrated with Google Earth’s satellite, aerial, and 3D-mapping data. Users could generate images by typing text prompts, such as depicting a nonexistent flood near the U.S. Capitol or a smoldering crater in place of the Eiffel Tower. While Google initially claimed all AI-generated images were watermarked with SynthID, critics argued the tool’s ease of use and realistic output posed significant risks to public trust in geospatial imagery.
Open-source intelligence experts and researchers were among the first to highlight the tool’s potential for abuse. Henk van Ess, an AI researcher, demonstrated the feature’s capabilities by generating fake refugee camps along the U.S.-Mexico border and a nuclear plant in Iran. Other users created AI images of a 9/11-style attack and a fictional explosion at Mar-a-Lago, which circulated widely online before Google intervened. The rapid spread of these images raised alarms among journalists and emergency managers, who rely on verified satellite data for crisis reporting.
Google had marketed the feature as a way to “visualize history, create real estate plans, and more,” but the company’s initial response to concerns was limited. In its first statement, Google directed users to check AI-generated images themselves, noting that SynthID watermarks could be verified via the Gemini app or Google Lens. However, the company reversed course after public backlash intensified, including criticism from visual forensics investigators and media organizations.
The reversal follows a pattern of tech companies quickly pulling AI features after public outcry. Meta, for example, recently launched and then disabled its Muse Image tool following similar concerns about deepfake generation. The episode underscores ongoing debates about the balance between innovation and the risks of AI-driven misinformation, particularly in tools that interact with real-world data.
Trust in satellite imagery has taken decades to build, said Ross Burley, executive director of the Centre for Information Resilience. “It could be irrevocably damaged overnight,” he warned. The incident has prompted discussions among geospatial professionals about the need for stricter controls on AI-generated content in mapping platforms.