Launch and immediate criticism
On Thursday 30 July Google released a new function in Google Earth that let users generate AI‑altered versions of real satellite images. The tool used the Nano Banana 2 generator and, according to a blog post by product manager Bryan Horowitz, was intended for creative purposes such as visualising completed construction projects or turning the ruins of Pompeii into a hyper‑realistic view of the city in 78 BC. Within hours, open‑source investigators and journalists warned that embedding generative AI into one of the most trusted sources of satellite imagery for verification created a dangerous instrument for misinformation. Eliot Higgins, founder of Bellingcat, posted on Bluesky that Google Earth, often used as a source for verifying photos and videos, had added a feature allowing AI editing of satellite images for unknown reasons. He shared a generated image of a giant golden statue of Donald Trump over the White House with a note that misuse could not happen.
How the feature worked and why it alarmed investigators
Independent investigator Henk van Ess wrote on X that a single sentence sufficed to place refugees at the Mexican border, a nuclear plant in Iran or a fatal crash in Amsterdam onto genuine Google Earth satellite backgrounds. He referenced an earlier case where an image of the US Fifth Fleet base in Manama, Bahrain, had been transformed via Gemini into a fake picture of an Iranian drone attack. The process, which previously required roughly six steps, could now be executed in seconds using the built‑in tool. A BBC journalist on X added sarcastically that the new AI image generation feature in Google Earth, one of the most reliable visual evidence sources for journalists and researchers, could not be misused.
Google’s response and rollback
Google initially pointed to its SynthID system, which automatically inserts digital watermarks into images created by its AI tools. In a post on X on 30 July the company said it took misinformation seriously and that every image produced by Nano Banana in Google Earth contained a SynthID watermark, verifiable via the Gemini app or Lens in search. Only one day after launch, on Friday 31 July, Google announced on X that it was pulling the feature. The statement acknowledged that geospatial experts had used the tool for a range of useful purposes, but that cases of rule‑breaking through sharing generated images had emerged. The company therefore temporarily withdrew the function to implement stronger safety barriers. Ars Technica subsequently confirmed that the feature was no longer available to Google Earth users.
Implications for AI safety and verification workflows
TechCrunch noted that the criticism is understandable, yet in the AI era most visual material on the web can be easily manipulated; expertise in Photoshop is no longer required to create a half‑credible falsehood, only access to an image generator sold by Google and many other firms. The episode shows how quickly the integration of generative AI into verification tools can undermine trust in visual evidence. For organisations deploying AI systems it underlines the need to model misuse risks in advance, implement robust watermarks and access controls, and maintain a ready procedure for rapid rollback of features that fail real‑world safety tests.