Google Earth’s AI feature fiasco and takedown | Explained

The story so far: A new AI feature launched by Google in late July remained active for around a day before it was pulled down by the company, amidst serious safety and misinformation concerns. The launch of the feature and its quick rollback showed how easily even casual internet users broke past Google’s enterprise-level safeguards in order to create deepfake satellite images, in spite of the risks they pose.What was the AI feature launched by Google?Put simply, users could open Google Earth—the company’s immersive satellite map of the planet—in order to zoom into a place, choose the “create image” option, and type whatever they wanted to see built on top of the original imagery.On July 30, Google Earth’s Product Manager Bryan Horowitz shared a blog post that invited users to “transform any place,” with AI, the process powered by the company’s image generation Nano Banana technology.Google suggested several wholesome use cases for the feature, such as re-imagining (not visualising) how modern-day landmarks looked in ancient times, as well as generating mock-ups of completed real estate projects, creating educational infographics, and giving real-life locations a fantasy feel. Teachers, architects, and urban planners were invited to get creative with the prompts, and Google shared examples as well.Why was the AI feature taken down?On July 31, the company posted an update to its launch blog post, confirming that the AI feature in Google Earth had been rolled back over possible policy violations by others.“We know that people uniquely trust Google Earth for a reliable view of the world. 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,” stated the update, adding, “It’s important to note that generated images didn’t appear in the main Google Earth experience for others to see and were watermarked as AI generated.”While the official notice did not blame specific parties or provide examples of the offending images, the social media platform X (formally Twitter) showed satellite deepfake screenshots that depicted terrorist attacks, human crises, industrial accidents, environmental destruction, and more.Some of these images were generated by casual users who managed to access the feature, while others were the result of safety tests run by researchers. The deepfakes ranged in quality from unrealistically violent to highly natural, sparking experts’ concern.Many users online criticised Google over the rollout of the feature, claiming that the potential for misuse of AI in satellite images far outweighed any benefits such a feature could have. Others questioned why the enterprise-level safeguards were so weak and easy to bypass when creating satellite image deepfakes that depicted violence.What are the risks of fake AI satellite images?Google defended its feature by highlighting that the AI-generated images did not appear as part of the public Google Earth experience. Even so, there are risks in allowing casual users to create AI satellite deepfakes for personal, commercial, or entertainment value.Watermarks can be easily cropped out of such media, and the average user may lack the time and knowledge to further investigate a suspicious-looking satellite image — especially if it is presented in a short-form news video that is shared on social media and designed to go viral.Satellite image deepfakes could also be used to run real estate scams by changing land or geographic details, to the buyer’s detriment. Furthermore, these images would not be appropriate to use in serious academic projects or scholarly works.At a higher level, satellite image deepfakes could be created to spread harmful misinformation about ongoing conflicts, humanitarian tragedies, or even natural disasters.This is not a hypothetical danger; during the Iran war in February and March this year, the Tehran Times account on X shared an AI image that appeared to show the aftermath of bombing a U.S. target. This was made with older satellite images, which helped researchers classify it as a fake.Furthermore, satellite images themselves are not a perfect source of evidence. Multiple regions across the world have not been accurately updated over time, either due to inaccessibility or security regulations that prevent detailed street-level mapping. India and Israel are examples of such countries. To integrate high-quality AI deepfake generation features into these still fallible yet necessary resources adds yet more layers of risk.The rollback of AI image generation in Google Earth shows that the world’s largest internet company is still not ready to address these basic risks.