Google Pulls 'Nano Banana' Earth AI Feature After Fabricated Satellite Images Go Viral

Google pulled the AI image-generation feature it had just added to Google Earth, nicknamed "Nano Banana," on July 31, 2026 — a single day after it launched. The tool let users zoom into any location on Google Earth's web version, type a prompt, and generate a photorealistic scene layered over real satellite and aerial imagery.

Within hours of launch, users demonstrated the tool could fabricate convincing but entirely fictional satellite scenes: a blast crater in Los Angeles, a flooded U.S. Capitol, a fire at Iran's Kharg Island oil terminal, refugee camps at the US-Mexico border, and a fabricated Russian ICBM silo — none of which existed.

Researchers Henk van Ess and Bellingcat's Jake Godin publicly warned that because the fabricated scenes were built on top of authentic Google Earth base imagery, they could be mistaken for genuine satellite evidence once cropped and shared on social media, stripped of any context that they were AI-generated.

Google's usual defense against this — SynthID watermarking baked into Gemini-generated content — didn't reliably hold up either; BBC Verify testing found the watermark could be bypassed in some cases, undercutting the one mechanism meant to flag the images as synthetic after the fact.

Google confirmed the rollback and said it would only bring the feature back once it has built stronger guardrails, without giving a timeline.

The episode is a clean example of why "photorealistic and grounded in real data" is a worse combination than either alone — a fabricated image with no real anchor is easier to dismiss than one seamlessly stitched onto verified satellite tiles. Any product pairing generative AI with authoritative source imagery needs provenance and watermarking to survive contact with the open internet before launch, not after.