TL;DR: Google added its Nano Banana image model to Google Earth, letting users type a prompt and alter satellite imagery anywhere on the planet. Open-source intelligence researchers warned it would accelerate disinformation. Within roughly a day, Google paused the feature after users shared generated images that appeared to breach its policies.
The objection was not that the images were convincing in isolation. It was where they were being made. Verification work has long depended on Google Earth as a stable, time-stamped baseline to test suspect pictures against.
Sam Stockwell, who researches at the Alan Turing Institute, put it directly. “Google Earth has served for decades as the dated reference layer journalists and investigators check suspicious imagery against; building a generative tool into that same benchmark corrupts the very measure used to debunk fakes,” he said.
Henk van Ess, an open-source investigator who published doctored Google Earth pictures on X, was blunter: “The fake is now being made inside the thing people use to check whether pictures are real.”
What users produced immediately
The demonstrations arrived within hours. Users posted pictures they said showed an invented crater in Gaza, Iranian nuclear plants, and a collision on a street in Amsterdam. One generated crowds and tents across an empty paved area at the Mexican border.
Brady Africk, an analyst with the American Enterprise Institute, identified what set this apart from earlier generators. “These fakes are anchored to real satellite images, removing some of the tell-tale signs of AI generation like mismatched terrain and scale,” he said.
Google has said pictures produced by the tool carry a watermark its Gemini model can identify, and that they were kept out of the main Google Earth view. Africk doubted that would be enough: “Few people check whether an image is AI-generated as they scroll past or share it.”
The rollback
Google confirmed on Friday it was withdrawing the capability. It said geospatial professionals had been putting the feature to a range of useful purposes, but that people were also circulating screenshots of output which looked to breach company policy. It did not specify which images, and said the pause would last while it builds stronger guardrails.
Looking forward
This is the second such retreat in a month. Meta shut down Muse Image, a feature that built pictures from public Instagram profiles, following broad criticism that included objections from a Hollywood union. The pattern suggests policy review is arriving after launch rather than before it. For UK organisations that rely on satellite imagery for verification — newsrooms, insurers, humanitarian bodies — the durable lesson is that a trusted reference layer stays trustworthy only while nobody is allowed to write to it.