Google Reverts Google Earth AI Image Generator After Deepfake Backlash
Google has disabled its new AI image generation feature in Google Earth after users quickly created convincing deepfakes of non-existent events, bypassing initial safeguards.
Google has rolled back a recently introduced AI image generation feature within its Google Earth web application following widespread user-generated deepfakes that depicted fabricated events. The tool, powered by Google's Nano Banana 2 model, leveraged real satellite imagery as a base, enabling the creation of highly realistic artificial scenes.
Launched on July 30th, the feature allowed users to generate images from text prompts, with Google initially suggesting educational and professional use cases such as bringing history to life or creating real estate plans. However, concerns about its potential for misuse were quickly realized as researchers and news outlets demonstrated its capacity to produce photorealistic satellite imagery of events and locations that never occurred.
Open source intelligence researcher Henk van Ess showcased the tool's ability to generate images of refugees at the Mexican border, a nuclear plant in Iran, a crash in Amsterdam, and a bomb cratered hospital in Gaza, none of which were refused by the system. Similarly, NPR created fake images of fires in Iran and a submerged Washington D.C., while the BBC produced visuals of a collapsed Eiffel Tower and the Great Pyramid of Giza being swallowed by a sinkhole.
Despite initial safeguards, the BBC's Verify service was able to circumvent them by manipulating basic AI prompts. Google acknowledged the issue in a statement, citing policy violations and stating, "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."
The company did not rule out a future reintroduction of the feature. User reactions expressed skepticism, with comments suggesting that development teams must have foreseen such misuse and that Google was out of touch with potential consequences. The tool's fallback safeguards included a SynthID watermark, designed to be an invisible signal detectable by Google's AI services, and the fact that generated images were not initially visible to other users within the main Google Earth experience.
However, the effectiveness of SynthID was questioned, with reports indicating that it could not be reliably detected by Gemini and that users could easily screenshot and share generated images without the watermark being apparent. This incident highlights the ongoing challenge of distinguishing authentic content from AI-generated fabrications, particularly when leveraging realistic base imagery.
The misuse of this tool strikes at the core of how satellite imagery is used in open-source intelligence and journalism, where it serves as a critical reference layer for verifying events. This comes at a time when public trust in online content is already eroding, with a significant majority of people finding it increasingly difficult to discern real from AI-generated material.
In light of these developments, users are advised to exercise critical media literacy, verifying the source and provenance of any alarming satellite imagery encountered online. Checking for named reporters, official wire services, and clear captions is crucial, and anonymous posts should be treated with skepticism. The practical advice is to assume that any dramatic image, especially from an unverified source, might be an AI-generated fabrication.