[REVISION HISTORY]

Google Earth AI image tool withdrawn after backlash

Updated 23 times since CLSTR started tracking revisions of this situation.

What changed

2026-08-03 04:02 UTC → 2026-08-03 07:14 UTC · added ~~removed~~

Google launched the Nano Banana 2 AI image‑generation feature, branded “Create Image,” in Google Earth on 30 July 2026. The tool let users type prompts to produce photorealistic visuals over satellite, aerial and 3‑D map layers, each output marked with an invisible SynthID watermark. Within hours, researchers and journalists demonstrated that the feature could fabricate convincing scenes – including a bomb crater in Los Angeles, a flooded U.S. Capitol, a nuclear plant in Iran, refugee caravans at the U.S.–Mexico border, a cratered hospital in Gaza, a collapsed Eiffel Tower, Russian tanks in Kyiv and massive sinkholes. Critics warned that such deepfakes could erode trust in satellite imagery, a key source for fact‑checking and humanitarian monitoring. Google said it had observed both useful applications and policy‑violating screenshots, and on 31 July it paused the feature pending stronger guardrails, emphasizing that AI‑generated images never appear in the main Earth experience and that the watermark could be bypassed. On 1 August the company reiterated the suspension, pledged stricter safeguards before any future re‑launch, and gave no timeline for reinstatement. On 2 August Google announced it would roll back the feature entirely, disabling it and adding a visible SynthID watermark while strengthening safety guardrails. The company stressed that protecting the platform’s credibility required a complete pull‑back and that any future re‑introduction would only occur after robust safety measures are in place. Open‑source intelligence analysts such as Henk Van Ess and Brady Africk, who had warned that high‑fidelity fake satellite imagery could undermine confidence in the platform, praised the rapid response. The tool will remain offline until ~~robust~~ those safeguards are ~~in place,~~ implemented, underscoring the challenge of protecting the credibility of a service widely used for real‑world verification.

Versions

Only revisions since CLSTR began indexing content versions appear here. Select a version to see what changed compared to the one before it.