Google pulls Earth AI feature after one day over misuse

Google spent years integrating generative AI across its product suite—Gemini in Google Docs, AI Overviews in Search, and most recently Nano Banana 2 in Google Earth. But the Earth experiment unraveled within hours of launch. On the day it rolled out, users began abusing the feature to generate taboo imagery, forcing Google to pull the product less than a day after release.

Nano Banana is an AI image generation tool built into Google's Gemini chatbot. Unlike earlier image generators, it was designed to let users edit their outputs: users could query the system with simple requests, and the AI would add, remove, or alter objects within an image while maintaining realistic visual features—akin to Photoshop powered by generative AI. When Google integrated Nano Banana into Google Earth, the vision was legitimate: users could visualize planning concepts or historical reconstructions by overlaying AI scenes onto real satellite, aerial, and 3D imagery. But within hours, users began generating fabricated disasters, false military installations, and conflict scenes anchored to real-world locations. Specific examples include fabricated images of a nuclear plant in Iran and refugees near the Mexico-US border.

Google had built safeguards into the rollout. Generated images were separated from authentic Google Earth imagery in the interface, watermarked to signal AI authorship, and embedded with SynthID, a cryptographic watermarking technology designed to allow computers to verify AI origin. A Google spokesperson told Fortune that "generated images didn't appear in the main Google Earth experience for others… and were watermarked as AI generated." However, the company also acknowledged to Fortune that it was rolling back the feature "while implementing stronger guardrails," recognizing that the generated imagery "violated company policies."

Experts argue the protections were inadequate. Emily Black, an NYU computer science and engineering professor, told Fortune that while Google's SynthID system makes fabricated images "detectable," recent research has shown that detection alone is insufficient. SynthID works invisibly to machine-reading systems but requires a visible watermark for human users to notice AI authorship. In the Google Earth case, only digital watermarks were applied; when images were screenshotted or recreated outside the platform, that metadata and watermarking would be lost. Black also pointed to her own research on user trust: her work on AI hallucinations and text-based systems has found that "run-of-the-mill disclosures are often not effective at dissuading users from following harmful advice." Even with explicit warnings like "AI chatbots can make mistakes," the presence of incorrect AI-generated information "significantly increases a user's reported likelihood of taking a harmful future action." Her research documents real-world harms, including poor investment decisions based on AI-hallucinated financial information.

The core issue is institutional credibility. Google Earth has functioned as a trusted archive of the physical world for nearly two decades. Users approach its imagery with an assumption that it corresponds to reality. By introducing AI-generated content into that historically authoritative platform, Google created conditions for both misuse and a credibility crisis affecting the entire platform. Henk Van Ess, an international expert in online research methods in AI, framed the risk sharply: an official viewing a genuine photograph of a genuine atrocity could now simply claim "AI"—not because the tool is needed to lie, but because its existence provides plausible deniability. Google has not committed to relaunching the feature and told Fortune it would not immediately comment on whether it would return after stronger guardrails are implemented.