Google pulls AI image tool from Google Earth just one day after launch
Users quickly showed the new feature could fabricate convincing fake satellite imagery, like refugees at the border or bomb damage near a hospital.
What happened: On Thursday, Google added a feature to Google Earth letting users type a text prompt and generate AI images — using its Nano Banana 2 model — layered onto real satellite and 3D maps. Google pitched it for visualizing historic sites or real estate projects. Within a day, researchers and journalists showed it could just as easily fabricate scenes like refugees near the US-Mexico border or a bomb crater beside a hospital in Gaza. By Friday, Google had rolled the feature back entirely.
Why it matters: Google Earth is widely treated as a trustworthy visual record — used by journalists, researchers, and fact-checkers to verify real-world events. A tool that can convincingly insert fake scenes into that same platform threatens to undercut exactly the kind of evidence people rely on to sort truth from fabrication, especially around sensitive topics like migration or war zones.
How it works, plainly: Google said every AI-generated image carried an invisible SynthID watermark and that harmful prompts were blocked. But testers found neither protection held up well: one researcher got the tool to generate sensitive fake imagery without pushback, and separately fooled an outside AI-detection tool with a generated video, showing the watermark didn't reliably flag the content as fake once shared elsewhere.
The rollout: Google says generated images never appeared in the shared, public Earth view — only on the creator's own screen — and that the fakes spreading online were screenshots. The company called this misuse a policy violation, not a feature failure, and says it's rebuilding guardrails before trying again. No timeline was given for when, or whether, the tool returns.