Google Pulls Google Earth AI Image Tool After Day-One Fakes

A generative-AI feature layered on top of a map that people rely on to verify what is actually happening in the world has a very short honeymoon. Google added its Nano Banana 2 image model to Google Earth on July 30, 2026, letting anyone zoom to a place, tap 'create image', and type a prompt to alter the satellite view. Less than 24 hours later, Google pulled the feature back, after users produced exactly the kind of images the reporting warned it would produce.

The list is bad. AI researcher Henk van Ess generated a nuclear power plant in Iran, a refugee camp on the US-Mexico border, and a fake hospital in Gaza with a bomb crater next to it. BBC Verify managed a collapsed Eiffel Tower, a sinkhole swallowing the Great Pyramid of Giza, and Russian tanks in Ukraine's capital. Other users produced a blast crater in Los Angeles, a flooded U.S. Capitol, and a fire on Iran's Kharg Island. Google's original pitch, as NPR reported, was blander: 'visualize history, create real estate plans, and more.'

What makes this more than a routine guardrail miss is that satellite imagery has done real work as a source of ground truth. Open-source investigators use it to verify events during conflicts, and reporters lean on it when nothing else is available. Brady Africk of the American Enterprise Institute warned the update 'makes it easier to generate convincing fake satellite imagery that can spread quickly online and mislead the public,' and the Atlantic Council's Digital Forensic Research Lab has reportedly already observed users fabricating military positions and relabeling buildings. Google's response points to a SynthID digital watermark on every generated image, but researchers demonstrated that detection tools can be fooled and moderation bypassed.

The honest caveat is that the reporting does not tell you how much internal red-teaming Google did before shipping, or whether the paused version will return with narrower prompts, geofencing of sensitive locations, or human review in the loop. It also does not tell you how many images generated during the short live window are already circulating outside the app with the credibility of the map baked in. The forward-looking part is less about Google and more about the rest of the stack, because provenance standards like SynthID just got a very public stress test, and verification teams and rival mapping platforms now have a clearer product spec for what 'authoritative imagery' has to defend against.