On July 30, 2026, Google integrated its AI image generator "Nano Banana 2" into Google Earth — users could type a text prompt and conjure fictional scenes on top of real satellite imagery, aerial photography, or 3D maps, then download and share them. Less than 48 hours later, the feature was pulled in a hurry.

The questions it left behind were not pulled with it: when AI-generated imagery is fused with real geographic coordinates, the boundary of fabrication disappears.

One-Click "Satellite Photos"

This integration pushed generative AI forward from "drawing a picture" to "drawing a picture with coordinates." Users could overlay fictional content onto real satellite basemaps — a collapsed Eiffel Tower, a giant sinkhole swallowing the Pyramid of Khufu, Russian tanks appearing on the streets of Kyiv.

BBC's fact-checking unit tested the feature on the day it launched and successfully generated multiple images convincing enough to pass for the real thing. The cases demonstrated by Henk van Ess, an expert on AI and disinformation, were even more specific: a nonexistent nuclear power plant inside Iran, a refugee camp on the US-Mexico border, a fake hospital in the Gaza Strip with a bomb crater sitting next to it.

The danger lies not in the image itself but in its coordinates. A "satellite photo of a nuclear power plant" without coordinates is just another piece of AI artwork; the moment it is bound to real geographic coordinates on Google Maps, it becomes "sourced evidence" — during the information chaos of wars and disasters, journalists and the public can barely tell whether an image comes from reliable satellite data or from a text prompt.

The Watermark Arms Race

Google did not go in defenseless: every piece of AI-generated content carried an invisible watermark, and anyone doubting an image could verify it through Gemini or the Lens feature in Google Search. BBC's testing found the mechanism worked for most generated images.

But the verification mechanism itself is part of the contest. BBC noted at the same time that some verification mechanisms could still be bypassed, and external detection tools sometimes failed to identify fake images. Watermarks can be added — and they can also be cropped, compressed, or re-photographed. Every image that has ever circulated on social media has been through exactly those processes.

Google removed the feature within 48 hours, citing the need to "strengthen safety protections." But from the standpoint of information warfare, those 48 hours had already completed a proof of concept: real basemap + AI overlay + coordinate anchoring is enough to mass-produce "fake news with a geographic location." Pulling the tool does not make the capability disappear — the same technology stack can reappear through another entry point at any time.

📝 The Proof-of-Concept Window

Forty-eight hours was enough to complete a full proof of concept: BBC's same-day testing produced multiple images convincing enough to pass for real, and the expert demonstrated coordinate-anchored cases including a nuclear power plant, a refugee camp, and a fake hospital. Removing the feature merely closes the entry point; the technology stack itself has not gone away — this is a textbook sample of "the irreversible diffusion of capability."

Plugging Into the Existing Cognitive Warfare

This thread forms a progression with the "triple-edit technique" recorded on existing pages: editing technique means partial tampering with authentic footage, while AI generation produces, from scratch, a piece of coordinate-bearing "reality." It also adds a new dimension to the "infrastructure abuse" framework — not only telecommunications and payment infrastructure, but the map service itself can become a generation platform for disinformation.

For fields that depend on satellite imagery to make judgments — war reporting, disaster assessment, arms-control verification — this means verification costs rise sharply. In the past, "having satellite imagery" was a credibility bonus; now it is merely the starting point of verification.

When AI-generated imagery is fused with real geographic coordinates, the boundary of fabrication disappears.