Google Pulls Down Feature for AI-Generating Fake Satellite Images on Google Earth After It’s Immediately Used to Fake Terror Attacks

Google has removed an AI image generation feature from Google Earth after researchers immediately weaponized it to fabricate realistic satellite imagery of terror attacks, natural disasters, and military strikes, raising urgent questions about whether the company can safely deploy generative AI tools that inherit the credibility of authoritative map data. The incident exposes a critical institutional risk: AI-generated disinformation anchored to real geographic coordinates is harder to debunk than standalone fakes, potentially undermining trust in satellite imagery as evidence in geopolitical analysis, conflict monitoring, and investment due diligence.

• Google added an image generation tool to Google Earth powered by its Nano Banana 2 model allowing fabrication of satellite imagery anywhere globally
• Open source intelligence analysts immediately generated fake refugee camps, nuclear plants in Iran, warzone scenes, and bomb craters in Gaza
• Google pulled the feature within hours after viral exposure, citing commitment to fighting misinformation but offering no technical safeguards explanation

• Hours Time between Google’s feature launch and removal after OSINT researcher exposed misinformation capability
• Multiple Types of false catastrophes generated in testing: droughts, wildfires, nuclear incidents, warfare, and terror attacks
• Zero Apparent guardrails or content filters protecting against malicious synthetic imagery prompts

Google on Thursday announced a new AI-powered feature embedded directly into the web version of Google Earth, allowing users to generate photorealistic satellite imagery of any location on the planet using text prompts.

The tool, built on the company’s Nano Banana 2 model, promised to let people visualize hypothetical scenarios, urban development projects, climate impacts, or infrastructure changes.

Within hours, open source intelligence researchers and security analysts had weaponized it to fabricate catastrophic false events at authentic geographic coordinates: refugee encampments near the Mexican border, a nuclear facility in Iran, bombed-out neighborhoods in Gaza, and military occupation of major U.S. landmarks. By Thursday evening, Google had pulled the feature offline.

OSINT Researchers Exposed the Feature’s Core Vulnerability in Real Time

Hank Van Ess, an established open source intelligence analyst, demonstrated the feature’s danger by using it to generate a hospital with bomb craters overlaid on Gaza, authentic terrain, authentic coordinates, synthetic catastrophe. His tests laid bare why this tool posed a uniquely credible form of misinformation.

Google had advertised that the AI “creates concepts grounded in the real world,” a seemingly innocuous marketing phrase that concealed the actual risk.

When an AI invents a disaster at real map coordinates, it doesn’t stand alone as an obvious fake; it inherits the visual authority of the underlying satellite data, adopting the same color palette, light angles, and resolution as the genuine imagery adjacent to it.

The forgery does not have to look convincing on its own. It inherits the credibility of the map it was born on.

Hank Van Ess, open source intelligence researcher

This distinction is critical for institutional investors and geopolitical analysts who depend on satellite imagery as primary evidence. A standalone AI-generated image of a nuclear plant explosion in Iran might circulate on social media and face immediate skepticism.

But the same image anchored to real coordinates, displayed within Google Earth’s interface alongside authentic surrounding terrain, carries the implicit endorsement of one of the world’s most trusted mapping platforms.

Bad-faith actors could weaponize such imagery to trigger financial market reactions, justify investment exits from conflict-adjacent regions, or fuel narratives that drive geopolitical decisions.

Google’s Minimal Safeguards Failed to Stop Obvious Weaponization Attempts

Testing revealed that the feature had almost no functional content filters. Requests to visualize rural Pennsylvania suffering drought-induced crop death and forest fires produced photorealistic imagery with towering smoke plumes. Prompts asking the tool to show the same town “on fire and full of soldiers” complied without hesitation.

Van Ess’s request to add a hospital with bomb craters in Gaza generated output. A prompt requesting imagery of the White House “in the middle of a warzone” with explicit instructions to add “blood” and “fallen soldiers” was executed.

The AI’s outputs were not uniformly perfect, human figures often appeared disproportionate to their surroundings, and some composite elements looked amateurish when examined closely. But Van Ess and other testers noted a critical truth: imperfect AI output is not the same as obvious fakery.

A user with time and technical skill could iteratively refine prompts until producing imagery sophisticated enough to deceive initial viewers, particularly in fast-moving social media environments where real-time verification is difficult. The low bar for deployment stood in sharp contrast to the high bar for public trust.

Google’s Rapid Removal Avoids Transparency on Why Safeguards Were Absent

After Van Ess’s findings went viral on social media, Google issued a brief statement acknowledging that it takes misinformation seriously but provided no explanation for why the feature had shipped with such minimal controls.

The company did not outline what specific guardrails it had attempted to implement, why those guardrails failed, or what technical or policy changes would prevent similar incidents.

The speed of removal, hours rather than days, suggested either that internal teams had not anticipated the risk, or that the company prioritized immediate damage control over institutional communication with researchers, security professionals, and policymakers who rely on Google Earth data.

The incident creates a lasting credibility problem for Google Earth as an institutional reference tool. Professional analysts in defense, intelligence, journalism, and investment already face skepticism when citing satellite imagery as evidence, particularly in politically contentious zones.

The knowledge that AI-generated imagery has been embedded in and then removed from Google Earth may now create persistent doubt about whether any given image is authentic, even if all future AI generation tools are disabled. Trust, once undermined, is not quickly restored by removal alone.

Institutional Investors Face New Risks Assessing Geopolitical Claims Using Visual Evidence

For institutional crypto investors and traditional finance players conducting due diligence on crypto infrastructure located in conflict zones, politically unstable regions, or jurisdictions with contested sovereignty, satellite imagery has become a key tool for assessing real-world risk.

Investors verify claims about mining operations, data centers, and financial infrastructure by cross-referencing satellite data. If public trust in the integrity of satellite imagery erodes, so does the reliability of one of the few available verification methods for on-the-ground asset claims.

A venture capital firm evaluating a blockchain infrastructure project in an emerging market may now reasonably ask whether recent satellite imagery can be trusted, a question that would have seemed absurd a week earlier.

Google has not disclosed whether the Nano Banana 2 model will be retained for use in other products, what specific training data or guardrails it intends to implement before any future redeployment, or whether it will submit the tool to external security review before release. The company also has not stated whether it will audit past Google Earth sessions to identify and flag any AI-generated imagery that may have been created and shared during the brief window the feature was live. Until Google provides concrete technical and policy commitments addressing how it will prevent AI-generated imagery from inheriting the credibility of authoritative maps, institutional users should assume that any satellite-based due diligence conducted via Google Earth may require independent verification through alternative platforms.