Google quickly withdrew a new AI image-generation feature in Google Earth after researchers demonstrated its disturbing ability to produce realistic satellite-based misinformation.
The feature, powered by Google’s Nano Banana 2 AI model, launched on July 30. It let users modify Google Earth’s authentic satellite, aerial, and 3D imagery with simple text prompts.
At first, Google described the tool as an educational and creative resource. The company suggested people could use it to imagine unbuilt real estate projects or see what Pompeii’s ruins looked like during the Roman Empire.
However, independent investigators quickly exposed the tool’s vulnerabilities. Open-source intelligence researchers, including Bellingcat founder Eliot Higgins and independent investigator Henk van Ess, highlighted how easily the feature could be weaponized to generate false scenarios overlaid on real-world locations.
READ MORE: Google says AI fixed 1,072 security bugs in Chrome across two recent releases
Van Ess showcased the tool’s risks by creating fake images of refugees at the US-Mexico border, an Iranian nuclear plant, and a fatal accident in Amsterdam—all based on Google’s authentic satellite maps.
The researchers cautioned that journalists rely heavily on Google Earth to verify global events and conflict zones. If users can easily add deceptive features—such as bomb craters or military strikes—to these trusted maps, it could significantly damage the platform’s credibility.
In response to severe backlash, Google announced on July 31 that it was rolling back the feature less than 24 hours after its debut. The company said it was pausing the capability to implement stronger guardrails, acknowledging that users were sharing screenshots of generated imagery that violated its policies.
Before the rollback, Google defended the tool by citing its SynthID digital watermarking system, meant to detect AI-generated content.
However, researchers disagreed, saying that bad actors rarely share original digital files. Instead, they usually spread screenshots, re-encoded videos, or photos taken from screens. These methods can easily get around digital watermarks, making verification almost impossible.
Google says it plans to rework the tool with improved safeguards, but for now, the Nano Banana 2 integration remains inaccessible to Google Earth users.