Google pulled the Nano Banana 2 image tool from Google Earth one day after shipping it, once a researcher demonstrated that a one-sentence prompt could drop a nuclear facility into Iran. The safety failure is the obvious story. The more interesting one is that nobody, before or since, has said what the feature was for.

Satellite imagery has always been an instrument. Insurers use it to assess claims. Commodity traders use it to count cars in car parks and read crop yields. Militaries use it for the reasons militaries have always used it. On the humanitarian side, SKAI — built by the same Google with the UN World Food Program — reads post-disaster imagery to assess building damage at scale. Every one of those is AI reading a picture.

There is no satellite photographer. Nobody frames the shot. Nobody waits for the light. The image is produced by a sensor on a schedule, and its entire value is that it is a measurement rather than a composition. A generative editing layer on top of that is a can opener invented before the can. Photoshop before cameras. Editing presupposes an author and something the author was trying to say, and satellite imagery has neither.

I live somewhere that treats mapping data as a strategic asset. Korea blocked the export of high-resolution survey data for nineteen years, and when it finally cleared Google this February, it attached conditions — military and sensitive facilities blurred, coordinates for Korean territory stripped from global products, raw data processed on domestic servers. The restriction regime did not go away. It issued a permit. The same logic reaches further than software: during the presidential office relocation, security rules restricted photography from nearby parks and mountains. That isn’t a quirk. It’s a reminder that the people who hold this imagery have never once regarded it as a creative medium.

Google’s own showcase example was Pompeii before the eruption. Take it seriously for a moment. As reading, it offers nothing — a generated guess at a buried city is not evidence, and no one studying the site would treat it as such. As writing, it offers nothing either. Anyone drawn to Pompeii wants to walk the streets, and the views that sell that are taken at eye level or from a drone at fifty meters. Orbit is where the sensor happens to be. It is not where anybody would stand.

Which raises the question of whether this was a capability at all. Fabricated Google Earth imagery was already circulating before the feature existed. During the US–Iran exchange, a state-aligned outlet posted a “before and after” it said showed destroyed US radar equipment in Qatar. Reporting traced it to an AI-manipulated Google Earth image of a base in Bahrain, captured a year earlier. Someone did that with general-purpose tools and a screenshot. No feature required.

So what shipped was prompt-based photo editing with a satellite skin, placed inside the one interface that lends the output institutional credibility. That credibility was the only thing the feature actually added. It is also the thing that got it pulled. As a showcase for Nano Banana 2, it showcased very little.