Google Disables AI Image Generation in Google Earth After Deepfake Exploits

Google has reversed course on an artificial intelligence feature in Google Earth that enabled users to edit satellite imagery using simple text prompts, just one day after its launch, The Verge reported. The tool, which relied on an AI model to generate photorealistic alterations, was intended for geospatial professionals but quickly drew criticism for its potential to create misleading or harmful deepfakes of real-world locations.

Security researcher Henk van Ess demonstrated the tool’s risks by generating images that added refugees near the Mexican border and a bomb crater next to a hospital in Gaza. Despite Google’s claims that all generated images were digitally watermarked and that safeguards blocked harmful topics, van Ess found that in his tests, “nothing was refused, nothing was softened, and nothing suggested I try a different prompt.” He also showed that a video created in Google Earth could fool the AI detection service Hive, casting doubt on the effectiveness of the watermarks.

Google initially responded to van Ess by reiterating that the feature prevented image creation on harmful subjects, but within a day the company issued a new statement announcing the rollback. “We know that people uniquely trust Google Earth for a reliable view of the world,” a Google spokesperson told The Verge. “We’ve seen geospatial professionals using this feature for a range of useful purposes, however we’ve also seen people sharing screenshots of generated imagery that appear to violate our policies. So we’re rolling back this feature in Google Earth while we work on implementing stronger guardrails.”

Google emphasized that the AI-generated images did not appear in the main Google Earth experience and were not visible to other users, but the fact that users could easily capture and share screenshots of the manipulated views amplified the risk of misinformation. The episode underscores the challenges tech companies face in deploying generative AI tools that alter real-world imagery, where the line between creative utility and deceptive misuse is thin.