Google's AI satellite imaging tool sparks disinformation fears among researchers

Google rolled out a new feature in Google Earth on Thursday that lets anyone with a browser zoom into a location and ask an AI to generate images based on satellite data. The tool, powered by the company's Nano Banana 2 image-generation system, works in seconds. Google framed it as a way to visualize history, sketch out real estate plans, explore possibilities. But within hours of the launch, researchers and open-source intelligence experts began raising alarms: the feature could become a factory for convincing geospatial lies.

The concern cuts deeper than typical AI-generated-image worries. Google Earth has spent two decades becoming the reference standard—the thing journalists, fact-checkers, and conflict monitors turn to when they need to verify what actually happened on the ground. A digital investigations expert named Henk Van Ess put it plainly on Substack: "Google spent 20 years building the reference the world checks against. Today it added a button that makes things up." The real danger, he noted, is that a fabricated image doesn't have to look perfect on its own. It inherits the credibility of the map it was born from. A fake satellite image of an explosion in Paris, posted alongside the real Google Earth layer, carries the weight of Google's authority.

When AFP tested the tool on Friday, it took only minutes to generate satellite imagery of an explosion in Paris, a nuclear facility in Iran, a bomb crater in Russia, and an Islamic State training ground in Syria. The system also produced fakes of floods across Bangladesh, a migrant caravan at the US-Mexico border, and a warehouse in Georgia—a state that former President Donald Trump has repeatedly and falsely accused of election fraud. Each image looked plausible enough to fool a casual viewer, and plausible enough to spread.

Google responded to the criticism by pointing to two safeguards. First, images created with the tool carry SynthID watermarks—invisible digital markers embedded in AI-generated media that can be detected by Google's Gemini chatbot. Second, the company said it prevents image creation on harmful topics and has policies against misinformation. But those guardrails did not stop AFP from generating the fakes. The watermarks exist, but they are invisible by design, and most people sharing an image online would have no idea how to check for them. The policy restrictions, meanwhile, proved porous in practice.

Brady Africk, a satellite imagery analyst at the American Enterprise Institute, told AFP the update makes it trivially easy to produce convincing fakes that spread fast and mislead the public. More troubling still, he said, is the long-term damage: as AI-generated satellite fakes become more common, people may stop trusting satellite imagery altogether. That erodes the work of journalists and researchers who depend on visual evidence to document what happened.

Ross Burley, executive director of the London-based Centre for Information Resilience, was blunt: "I cannot stress how irresponsible this is. Trust in satellite imagery has taken decades to build and could be irrevocably damaged overnight." His organization is calling on Google to reverse the decision. GeoConfirmed, a volunteer-run open-source intelligence project that verifies images from conflict zones around the world, reported on X that people are already using the tool to fabricate military positions, relabel buildings as schools, and test out ways to manipulate the information environment. "The primary, and perhaps only, practical use of this tool appears to be the creation and dissemination of mis- and disinformation," the group wrote.

This is not theoretical. In 2023, a single AI image of an explosion at the Pentagon briefly rattled financial markets. Early in the US conflict with Iran, an AI-altered satellite image of a destroyed military base racked up millions of views across social platforms. Now Google has handed millions of people the ability to generate similar fakes in real time, anchored to the most trusted geospatial database on Earth. The question is not whether the tool will be misused. It already has been. The question is how much damage it will do before anyone can stop it.