The numbers come first. Google's new AI image-editing feature for Google Earth survived less than one full day before the company shut it down — a timeline that says everything about the gap between a product launch and a product ready for the public.
The feature, powered by Google's 'Nano Banana 2' technology built into its Gemini chatbot, allowed users to overlay AI-generated scenes onto real satellite, aerial and 3D imagery. Google envisioned legitimate uses: urban planning visualizations, historical reconstructions. What it got instead was fabricated disaster scenes, false military installations and manipulated imagery tied to real geographic coordinates — including generated images of a nuclear plant in Iran and refugees near the Mexico-U.S. border.
Google told Fortune it was rolling back the feature 'while implementing stronger guardrails,' acknowledging that the generated imagery appeared to violate company policies. A company spokesperson also noted that 'generated images didn't appear in the main Google Earth experience for others… and were watermarked as AI generated.'
The watermark defense, however, is where experts push back hardest. Google deployed two layers of identification: visible watermarks on the generated images and 'SynthID,' a cryptographic watermarking technology embedded in the image metadata. But NYU computer science and engineering professor Emily Black told Fortune that SynthID is 'invisible to the naked eye' — a machine-readable tag, not a user-visible flag. Any screenshot or image recreated outside Google's platform would strip the original metadata entirely, leaving no trace of its synthetic origin.
'To their credit, Google has a system that makes fabricated images detectable,' Black told Fortune. 'Even if there were, however, recent research has shown that's not a full solution.'
Black's own research on AI hallucinations adds a deeper layer of concern. Even when users are explicitly told content is AI-generated, disclosure alone does not reliably prevent them from trusting it. 'Even with disclaimers like, 'AI chatbots can make mistakes. Check important information,' the presence of incorrect AI-generated advice significantly increases a user's reported likelihood of taking a harmful future action,' she said.
The European Union's AI Act already requires providers to make synthetic content detectable — a transparency standard Google addressed through SynthID. But the professor argues the policy architecture has not kept pace with the technology's capacity for harm, calling on policymakers to collaborate with technical experts on stronger solutions.
Google Earth is not a generic creative canvas. For nearly two decades it has functioned as a trusted visual record of the physical world — a reference tool for journalists, researchers, governments and ordinary users who treat what they see there as ground truth. That institutional credibility is precisely what made the Nano Banana integration so dangerous, and why the disinformation potential was orders of magnitude higher than in a standard AI image generator.
The market has already voted on what happens when a platform confuses brand equity with a testing environment. Google's decision to launch with digital-only watermarks — no overlay visible to the human eye — was a safeguard built for regulatory compliance, not for real-world abuse. The rollback is the right call. The lesson is that trust, once weaponized, is very hard to restore. Free enterprise rewards innovation, but it also punishes companies that ship products that turn their own reputation into a liability. Google now owns both the launch and the cleanup.
