Google has withdrawn a new artificial intelligence feature from Google Earth only one day after its release, following the circulation of generated images that appeared to violate the company’s policies.
Alphabet-owned Google rolled back the image-generation capability on July 31 after users shared screenshots of AI-created scenes based on real-world locations.
Bloomberg described the reversal as a response to concerns about fake images produced through the Google Earth tool.
Nano Banana 2 Altered Real-World Locations
According to Reuters, the feature used Google’s Nano Banana 2 model to produce photorealistic images grounded in Google Earth’s satellite, aerial and three-dimensional imagery.
Users could type a text prompt for any location and instruct the system to create a new visual scene over the selected place.
The close connection between generated content and recognizable geographic imagery created an unusual misinformation risk. Unlike an AI picture produced on an empty canvas, an altered Google Earth image could retain familiar roads, buildings, monuments or surrounding landscapes while introducing an event that never occurred.
Critics were concerned users could create false scenarios involving real monuments and landmarks, making fictional events appear connected to authentic geographic evidence.
Google Cites Screenshots That Violated Policies
Google acknowledged that the feature had legitimate applications but said its early use also revealed problems requiring stronger safeguards.
In a statement posted on X, Google said geospatial professionals had used the capability for “a range of useful purposes,” while other people circulated generated screenshots that appeared to breach its policies.
Google paused the tool while it develops stronger guardrails, although the company did not identify the specific categories of imagery that triggered the rollback.
The lack of detail leaves unanswered questions about whether users bypassed existing restrictions, produced prohibited violent or deceptive material, or simply demonstrated that the tool’s current protections were insufficient for realistic geographic imagery.
Watermarks Did Not Eliminate the Risk
Reuters said the generated images carried labels identifying them as AI-created and were not displayed to other users through the main Google Earth experience.
Those protections reduced the likelihood that fabricated scenes would permanently replace official Google Earth imagery. However, screenshots could still be removed from their original context and distributed through social media, messaging applications or other platforms.
That distinction helps explain why visible labeling alone may not resolve the problem. A watermark can identify synthetic content when it remains intact and visible, but it cannot prevent users from misunderstanding, cropping or deliberately misrepresenting the image after it leaves the original product.
AI Rollouts Face Faster Public Testing
The Google Earth reversal reflects the difficulty technology companies face when adding generative AI to widely trusted products. Tools may be designed for planning, education or visualization, yet public access can quickly expose uses that were not anticipated during internal testing.
Meta discontinued its Muse Image feature earlier in July after the tool, which generated images using public Instagram accounts, attracted widespread privacy criticism that included objections from a Hollywood union.
Google has not said when the Google Earth capability could return. Its brief release nevertheless demonstrates that combining generative AI with realistic geographic data requires more than labeling the output. The safeguards must also account for how convincingly fabricated scenes can imitate visual evidence once they are shared beyond the platform.
