Google Pulls Google Earth AI Feature One Day After Launch Over Fake Satellite Images
Launched July 30, the feature let users generate AI imagery tied to real locations — but researchers showed it could convincingly fake a nuclear facility in Iran and a bomb crater near a hospital in Gaza.
Nova AI News Editor
August 10, 2026 · 2 min read
Google rolled back its new Google Earth AI feature on July 31, 2026, just one day after launching it, citing the risk of spreading misinformation. The feature let users generate AI images tied to real geographic locations using actual Google Earth imagery and text prompts.
What exactly went wrong?
Researchers quickly demonstrated that the tool could produce extremely convincing fake satellite images. Examples included a fabricated bomb crater next to a hospital in Gaza and a realistic-looking nuclear facility in Iran that doesn't actually exist. Even more concerning, these fake images passed Google's own watermark verification system — meaning the tool failed to reliably flag its own AI-generated output as such.
Why this is particularly dangerous
Satellite imagery carries a special status as "objective evidence" in journalism, human rights investigations, military intelligence, and public debate. People routinely turn to satellite images to verify whether an attack actually happened somewhere, or whether a facility actually exists. The ability to convincingly fake this kind of imagery — to the point of being indistinguishable from the real thing — has serious implications ranging from disinformation campaigns in conflict zones to full-blown international incidents.
How did Google respond?
Google acknowledged that some shared images violated its policies and said it would not relaunch the feature without stronger safeguards. The company said it plans to bring the feature back with more robust digital markers and content moderation, but gave no firm timeline.
What this means for the industry
The episode is a clear signal that combining generative AI with "real-world data" — maps, satellite imagery, official documents — carries far higher risk than plain text generation. A chatbot phrasing a sentence wrong isn't in the same category as a map convincingly depicting a military facility that doesn't exist. Google's one-day rollback shows that even a major tech company didn't adequately stress-test the misinformation risk before shipping — and similar tools should expect much stricter scrutiny going forward.
Comments
No comments yet — be the first to comment.