Google Earth spent two decades earning a kind of trust most tech products never get. Journalists, war-crimes investigators, and fact-checkers around the world treat it as close to a neutral record of what’s actually happening on the ground. On July 30, 2026, that record got a lot less neutral. For about a day. Google’s new Google Earth AI feature let anyone type a sentence and generate a fabricated scene laid directly over real satellite coordinates. Within hours, testers had used it to fake nuclear facilities, bomb craters, and refugee camps that never existed. Google pulled the tool less than 24 hours after launch. Here’s what happened, why researchers reacted so fast, and what it means for how much you can still trust a satellite image.

What Was Google’s Google Earth AI Feature Supposed to Do?

Google framed this as a creative add-on, not a change to the map itself. The “create image” button, added to Google Earth’s web version, let users zoom to any location, type a prompt, and generate a photorealistic image built from that spot’s real satellite, aerial, and 3D data. The pitch was equal parts nostalgic and practical. Google suggested picturing a city block the way it looked a century ago, or turning a vacant lot into a mocked-up community garden. The company’s marketing pointed to real estate concepts, historical visualizations, and classroom-ready graphics as the intended use cases.

Where Nano Banana 2 Fits In

The feature ran on Nano Banana 2, the newest version of Google’s Gemini-powered image generator. The original Nano Banana went viral after its 2025 debut, racking up billions of generated images within months. Folding that same model into Google Earth’s satellite data turned out to be a very different kind of test.

Why Did Google Pull the Feature So Fast?

The rollback took less than 24 hours, and it wasn’t entirely Google’s own call. Tech outlet 404 Media published tests showing the tool adding skyscrapers to farmland, a fake homeless encampment to a real Los Angeles neighborhood, and a fabricated bomb crater onto city streets. Other newsrooms ran their own tests within hours and got similarly convincing results. Google didn’t dispute what the testing found. In a statement posted to X and shared with the wire service AFP, the company said it had seen geospatial professionals use the tool productively, but had also seen generated images shared that broke its rules. It added only that it was working on “stronger guardrails” before bringing the feature back. Google also clarified one technical detail: none of the fabricated images touched the shared, public version of Google Earth that other users browse. Anything generated stayed on the creator’s own screen rather than the base map layer everyone else sees.

What Kind of Fake Images Did People Actually Create?

Once word got out, testers across multiple newsrooms raced to see how far the tool could be pushed. Within a day, reported examples included:

• A nuclear facility fabricated inside Iran
• A bomb crater dropped into a Russian city
• An explosion generated in the middle of Paris
• A staged militant training camp in Syria
• A migrant encampment placed at the US-Mexico border
• The Eiffel Tower shown toppled over
• A sinkhole swallowing the Great Pyramid of Giza
• Russian tanks inserted into central Kyiv
• A flooded US Capitol building
• A plane depicted crashing into a skyscraper A few tests leaned surreal instead of dangerous, including one showing a movie-monster-style creature flattening a small town. That range was itself the problem: nothing in the tool distinguished an obvious joke from a fabrication built to deceive. Both came out looking equally convincing.

Why Does This Matter So Much on Google Earth Specifically?

AI image generators have been able to fake scenes for years. What alarmed researchers here was the platform, not the technology. Google Earth is one of the few tools journalists, human rights investigators, and open-source intelligence, or OSINT, analysts treat as close to ground truth, precisely because faking a satellite photo has historically been slow and expensive. Digital investigations expert Henk Van Ess put it bluntly: “Google spent 20 years building the reference the world checks against.” He also flagged an asymmetry worth sitting with. Getting Google to remove one real building from Earth typically takes a formal government request, while adding a fake one took a single typed sentence. Ross Burley, executive director of the London-based Centre for Information Resilience, called the rollout irresponsible. He warned that trust in satellite imagery “has taken decades to build and could be irrevocably damaged overnight.” GeoConfirmed, a volunteer network that verifies conflict imagery for journalists and researchers, said its own testers had already found people using the tool to mock up fake troop and equipment positions before Google shut it down. A Bellingcat researcher made a related point to NPR. The risk was never that fabricated satellite images could exist, it’s that this tool made producing them fast enough to spread at scale, more like a deepfake generator wearing a mapping app’s credibility.

Is a Watermark Enough to Stop Misuse?

Google’s main safeguard was SynthID, an invisible digital watermark embedded in every image the tool generated. In theory, anyone unsure about a picture could run it through the Gemini app or Google Lens to check whether it was AI-made. Critics said that solution misunderstands how misinformation actually spreads. A watermark only helps someone who already suspects an image is fake and takes the extra step to check it. Viral images rarely get that kind of scrutiny before they’re screenshotted, cropped, and reshared without context. Brady Africk, a satellite imagery analyst at the American Enterprise Institute, told AFP that the bigger danger was speed. Realistic fakes could spread online long before anyone thought to check them.

This Isn’t the First Fake Satellite-Style Image to Cause Real Damage

Fabricated crisis imagery has caused real problems before. In 2023, an AI-generated image of an explosion at the Pentagon spread widely enough to briefly move US stock prices before it was debunked. During last year’s brief conflict between the US and Iran, an altered satellite image claiming to show a destroyed American base picked up millions of views before it could be confirmed as fake. Neither of those images came from inside a platform as widely trusted as Google Earth. That’s the shift researchers are worried about. Not that fake satellite imagery can exist, but that it might soon carry an implicit stamp of Google’s reliability.

Will the Google Earth AI Feature Come Back?

Google has framed this as a pause, not a cancellation. The company says it’s building stronger safeguards before reintroducing image generation to Google Earth, though it hasn’t shared a timeline or specifics on what those guardrails will actually involve. That leaves an open question for anyone who found a legitimate use for the tool, including urban planners, architects, and educators who wanted quick visual concepts tied to real locations. It also leaves open whether any set of guardrails can close the gap between a useful creative tool and a disinformation generator, given that the underlying issue wasn’t really a bug. The tool did exactly what it was built to do.

How Can You Tell If a Google Earth Image Is AI-Generated?

A few habits help, even though none of them are foolproof.

• Run a suspicious image through Google Lens or the Gemini app to check for a SynthID watermark
• Treat any viral satellite image tied to a breaking disaster, conflict, or election with extra skepticism until a named news outlet confirms it
• Remember that Google says generated images never appeared on the shared Google Earth map layer, so anything claiming to be a live Earth view showing a dramatic event is worth doubting by default
• Cross-check dramatic imagery against multiple independent sources before sharing it further

The Bottom Line

Google’s Google Earth AI feature lasted about a day, but the questions it raised will stick around much longer. The company built a genuinely capable image generator, then placed it inside one of the few digital tools people treat as objective. That combination, more than any single fake image, is what worried the researchers who pushed Google to pull it. Whatever guardrails Google builds next will have to answer a harder question than whether the AI can generate a convincing fake. It’s whether any watermark can catch up to a screenshot before that screenshot goes viral.

Tldr

• Google added a “create image” button to Google Earth on July 30, 2026, using its Nano Banana 2 model to generate AI images over real satellite, aerial, and 3D map data
• Testers quickly used it to fabricate convincing scenes at real coordinates, including nuclear sites, bomb craters, toppled landmarks, and refugee encampments
• Google rolled the feature back the next day, saying some generated images appeared to violate its own policies
• Researchers said the bigger risk wasn’t the AI itself, but where Google placed it, inside a platform millions treat as verified evidence
• Google says the feature may return once it builds stronger safeguards, though it hasn’t given a timeline