Rollback

Google rolled back its Google Earth AI image generation feature on Friday, less than 24 hours after launching it. The "create image" tool, which integrated Nano Banana 2 image-generation technology into Google Earth, allowed users to generate AI visualizations on top of satellite imagery. Following a furious backlash from researchers and open-source intelligence experts, Google paused the feature.[1]

"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," a Google spokesperson said in a statement. "So we're rolling back this feature in Google Earth while we work on implementing stronger guardrails." The company added: "We know that people uniquely trust Google Earth for a reliable view of the world."[2]

AFP's tests on Friday had already demonstrated the problem. Using the tool, they fabricated satellite images that could have had serious global consequences: an explosion in Paris, a nuclear site in Iran, a bomb crater in Russia, and a training ground in Syria. The tool also generated fake scenes of floods in Bangladesh, a migrant caravan at the US-Mexico border, and a suspicious warehouse in Georgia. These are not hypothetical risks. They are the actual outputs of a tool that was live for less than a day.[3]

The speed of the rollback is worth noting. Thursday: launch. Friday: rollback. That is not a deliberative process. That is a company reacting to immediate, unanimous, and well-sourced backlash from the exact community that would have to deal with the consequences. The Centre for Information Resilience called the feature "irresponsible." A volunteer open-source intelligence project, GeoConfirmed, reported that individuals were already exploiting the tool to fabricate military positions and relabel buildings. Their assessment was blunt: "The primary, and perhaps only, practical use of this tool appears to be the creation and dissemination of mis- and disinformation."[4]

Yesterday I wrote about the launch. The quote that captured the problem was: "Google spent 20 years building the reference the world checks against. Today it added a button that makes things up." The rollback does not undo the damage. The tool was live for roughly 24 hours. In that time, people generated fake satellite imagery of conflict zones, shared it online, and demonstrated that the guardrails, the SynthID watermarks, the policy instructions, the "we take misinformation seriously" statements, did not prevent the most obvious abuses. The watermarks are a detection mechanism, not a prevention mechanism, and they require the viewer to know they exist and to check for them. Most people seeing a satellite image on social media do not run it through Gemini first.[5]

Google says it will work on implementing stronger guardrails. That is the right response. It is also the response that should have come before the launch, not after. The pattern is consistent: ship the capability, discover the abuse, roll it back, promise to do better. The AI industry has repeated this cycle enough times that "stronger guardrails" is no longer a reassurance. It is a prediction of what will happen next time, after the next feature, after the next 24 hours of damage.[6]