Google Quickly Rolls Back a Feature That Could Make Deepfake Satellite Images

Google swiftly withdrew a new feature that could facilitate the creation of convincing deepfake satellite images on Thursday. Initially introduced the previous day, the Nano Banana 2 AI model allowed users to generate images based on Google Earth's satellite, aerial and 3D imagery. Concerns were swiftly raised by journalists and researchers over the potential misuse of this tool for misinformation campaigns.

The backlash prompted Google to announce it was turning back to the drawing board to enhance its safeguards. The feature, which allowed users to create custom infographics, reimagine past scenes, or envision future developments, was limited to desktop use and took a few minutes to generate. Users could refine and save these images to a project.

However, the tool's capabilities were quickly showcased in a viral post by researcher Henk van Ess, who demonstrated how easily it could produce fake but plausible images, such as a nuclear plant in Iran. Google initially described the feature as generating concepts "grounded in the real world." Following the criticism, Google announced on Friday that it would pull the feature after observing screenshots of policy-violating images.

The company stated that these images were not visible to others and were watermarked as AI-generated. While the rollback might provide some relief, it's important to note that the false reality could still be shared through screenshots or saved projects, potentially spreading misinformation and skewing public opinion.