Google pulls AI image-editing feature from Earth after one day of misuse

Google has disabled an experimental AI image-editing feature it introduced in Google Earth just 24 hours earlier. The capability allowed users to manipulate satellite photographs by typing text descriptions, effectively enabling the creation of convincing deepfakes of real-world locations.

Swift reversal amid misuse#

The company launched the feature Thursday, powered by its Nano Banana 2 model, with built-in safeguards including digital watermarks and restrictions on generating imagery depicting harmful content. However, within a single day, researchers demonstrated the tool's dangerous potential.

A Digital Digging contributor created several problematic images using the feature—including fabricated scenes of refugees at the Mexico-U.S. border and a bomb crater beside a hospital in Gaza. This raised immediate red flags about the technology's capacity to spread misinformation at scale.

Why the guardrails failed#

In a Friday statement, Google acknowledged that users had generated and shared screenshots of images that breached its content policies. The researcher also discovered that Hive, a prominent AI detection system, failed to identify video generated through the tool as synthetic media. He reported that his testing revealed the system imposed no restrictions: "Nothing was refused, nothing was softened, and nothing suggested I try a different prompt."

Next steps#

Google emphasized that the generated images remained watermarked as AI-created content and never appeared in other users' main Google Earth views. Still, the company determined the approach required stronger enforcement mechanisms before re-launching. The rollback highlights the tension between enabling useful applications—geospatial professionals had found legitimate uses for the feature—and preventing coordinated disinformation campaigns.