Why Google Earth’s AI Image Feature Was Pulled in Less Than a Day

On July 30, 2026, Google launched a "Create image" feature in Google Earth that allowed users to generate AI visuals layered directly over real satellite, aerial, and 3D imagery using its Nano Banana 2 model . Within roughly 24 hours, researchers and journalists demonstrated how easily the tool could fabricate convincing geospatial disinformation, and Google announced it was rolling back the feature on July 31, stating, “We are rolling back this feature in Google Earth while we work on implementing stronger guardrails” .

What the Feature Did

Google Earth’s “Create image” tool integrated Nano Banana 2, the company's image generation model, directly into the browser version of Google Earth. Users could zoom into any location on the globe, tap “create image,” and type a text prompt to generate a photorealistic scene grounded in Google Earth’s verified satellite and 3D mapping data . Google positioned the feature as a way to “visualize history, create real estate plans and more” . But the same capability that allowed a user to imagine a historic scene allowed another to simulate a bombing raid.

The Misinformation That Triggered the Rollback

Researchers and journalists moved quickly to test the feature’s limits. Their findings were stark.

• BBC Verify created a collapsed Eiffel Tower, a sinkhole swallowing the Great Pyramid of Giza, and Russian tanks rolling through Kyiv .
• The Atlantic reporter Matteo Wong generated office buildings on fire, mass destruction in San Francisco, a smoldering crater where the Eiffel Tower stood, a plane striking One World Trade Center, a homeless encampment on the White House lawn, protesters outside Mar-a-Lago, immigrants streaming over the U.S.-Mexico border, and a Trump Hotel on the Gaza Strip .
• PC Gamer reporters turned their hometowns into warzones with AI prompts .
• AI and misinformation expert Henk van Ess produced fake images of a non-existent nuclear power plant in Iran, a refugee camp on the U.S.-Mexico border, and a fake hospital in Gaza with a bomb crater next to it .
• NPR generated an image of Iran’s Kharg Island appearing to show damage .
• 404 Media’s Joseph Cox made an image of a blast using the tool .

Why Experts Called It Uniquely Dangerous

Open-source intelligence (OSINT) investigators and geospatial analysts, who rely on satellite imagery to verify wars, disasters, and human rights abuses, raised an immediate alarm. The core problem was that the tool did not create generic deepfakes. It anchored AI-generated fabrications to real geographic coordinates and real satellite maps.

Henk van Ess told BBC Verify that the tool “welded invented imagery to genuine coordinates, drawn on genuine imagery,” and that “the forgery does not have to look convincing on its own. It inherits the credibility of the map it was born on” .

Brady Africk, a research analyst at the American Enterprise Institute, noted that these fakes were “anchored to real satellite images, removing some of the tell-tale signs of AI generation like mismatched terrain and scale” . He added, “Few people check whether an image is AI-generated as they scroll past or share it” .

Investigative reporter Ben Heubl wrote that the tool could easily create images of drone strikes that look “very close to what we've seen in Ukraine, Russia, the Middle East, and other war-torn conflict zones” .

Henry Ajder, an AI-detection researcher, warned that fake imagery in fast-moving conflict zones could be particularly “destabilising,” because “spaces assumed to be unblemished by AI are valuable, particularly for journalists and people trying to do work on the frontlines of rapidly evolving crisis scenarios” .

Google's Safeguards—and How They Failed

Google had put two primary safeguards in place. First, every AI-generated image created with the tool included a SynthID digital watermark, and users could check whether an image was AI-generated by asking the Gemini app or using Google Lens . Second, the company built content guidelines into the AI model meant to prevent imagery depicting “harmful topics” .

In practice, both safeguards proved insufficient.

• BBC Verify found that while the SynthID watermark generally worked, it was possible to circumvent the checks and trick Gemini into saying fake Google Earth images were real. External AI detection tools also sometimes failed to identify the Google Earth AI content .
• Small changes to prompts could bypass the content guidelines. For example, a request for a “raised platform to hang traitors” near Parliament was rejected, but a less specific prompt was accepted .
• The watermark only works if users check it. Images can be downloaded and shared on other platforms where the watermark is easily stripped or ignored—a problem that critics said was obvious from the start .

The Broader Blow to Trust in Satellite Imagery

The episode exposed a deeper vulnerability. For decades, satellite imagery has been treated as one of the most trustworthy forms of evidence in journalism, war crimes investigations, and humanitarian work. AI-generated fakes that are anchored to real maps threaten that foundation.

Bill Greer, a geospatial analyst and co-founder of Common Space, told BBC Verify: “This is potentially particularly damaging to public trust because satellite imagery has historically been seen as an especially trustworthy source of data” . He noted that falsified satellite imagery has already been used to spread misinformation, and with governments restricting data access, “we're likely to see more false images, and fewer real images which will further erode trust” .

Henry Ajder added that the feature represented “another place where sources of trusted information are being eroded by the constant doubt that AI might be involved” .

Google's Statement and What Comes Next

On July 31, 2026, Google announced the rollback in a statement posted to X, saying:

“We've seen geospatial professionals using this feature for a range of useful purposes, however we've also seen people sharing screenshots of generated imagery that appear to violate our policies. We know that people uniquely trust Google Earth for a reliable view of the world. We are rolling back this feature in Google Earth while we work on implementing stronger guardrails.”

Google has not announced a timeline for the feature’s return or specified what stronger guardrails would look like. The incident remains a cautionary case study in how quickly a well-intentioned AI product can erode trust in a platform that journalists, investigators, and the public depend on as a source of ground truth.