AIBRIEF
Google Earth is the surface journalists use to verify real events. For 24 hours it also let anyone generate plausible satellite views; the company pulled the feature Friday.
Google pulled a "create image" feature from Google Earth on Friday, one day after adding it, after researchers warned the new AI image button could be used to forge satellite-style views on a product that fact-checkers and open-source intelligence researchers use to verify real events.
The feature, unveiled Thursday and built on Google's Nano Banana 2 image-generation model, let users generate plausible satellite-style images from the product's own satellite, aerial, and 3D-mapping data. AFP's own testing of the feature on Friday produced fabricated satellite views that looked verification-grade. A Google spokesperson told AFP the company had seen "screenshots of generated imagery that appear to violate our policies" and would add "stronger guardrails" before any relaunch.
Google's first line of defense was SynthID, an invisible watermark on generated images. Watermarks detect fakes after the fact. A verification surface needs a different mechanism: visible provenance on every generated view, in-product detection so a generated image cannot be mistaken for a basemap capture, and an explicit split between the generative layer and the authoritative satellite, aerial, and 3D basemap. "Stronger guardrails," the company's phrase, is now a separation problem rather than a watermarking one.
Henk Van Ess called the trust damage "irrevocable" overnight. Brady Africk told AFP the feature would let bad actors spread convincing fakes and "erode public trust in satellite imagery."