post-thumb

Google Earth’s AI deepfake tool lasted one day

Google has withdrawn an artificial intelligence feature from Google Earth one day after its launch, following concerns that users could create altered satellite images of real locations through text prompts. The tool allowed people to add fictional objects, events, and people to geospatial imagery, effectively producing synthetic versions of recognizable places.

Digital Digging’s Henk van Ess tested the feature by generating scenes that included refugees near the United States-Mexico border, a bomb crater beside a hospital in Gaza, and other fabricated elements. His demonstrations highlighted the possibility that convincing images could be detached from their original context and circulated as authentic evidence.

Google initially responded that images made with its Nano Banana 2 model carried digital watermarks and that safeguards blocked image creation involving harmful subjects. However, van Ess reported that his prompts were neither rejected nor modified, and that the system did not direct him toward safer alternatives during testing.

He also said a video created with Google Earth successfully bypassed detection by Hive’s AI analysis tool. That result raised questions about whether watermarks and automated detection systems would reliably identify altered material after it was recorded, reformatted, or shared through other platforms.

In a subsequent statement, Google said people place particular trust in Google Earth as a dependable representation of the world. The company acknowledged that geospatial professionals had used the feature for legitimate purposes, but said screenshots of generated imagery appeared to violate its policies.

Google therefore rolled back the tool while it develops stronger safeguards. The company emphasized that modified images were watermarked as AI-generated and were not displayed publicly in the main Google Earth experience for other users.

The episode illustrates challenges facing platforms that combine trusted mapping products with generative AI, particularly when synthetic content can resemble documentary imagery and spread beyond its source.

Share: