On July 30, 2026, Google connected its Nano Banana 2 image generator to Google Earth, enabling users to type a text prompt and have the AI paint new elements directly onto real satellite imagery. Google presented the capability as useful for city planning and historical illustrations.
Evidence from researchers
OSINT researcher Henk van Ess used simple, ordinary prompts — without jailbreaks or special exploits — to produce fabricated scenes: a nonexistent nuclear facility inside Iran, a bomb crater beside a Gaza hospital, and refugees at the U.S.–Mexico border. According to reports, the system accepted these requests rather than blocking them.
Google’s response
Google removed the feature in under a day.
Why this matters
For about two decades Google Earth has been treated as a reliable visual record: journalists, human rights organizations, and courts have used it for verification and to expose falsified satellite imagery. By making it possible to add AI-generated objects on top of authentic satellite photos, the company temporarily gave the same platform the power to manufacture the very deceptions it had helped to detect. Because the added content sits on top of real streets, angles, and lighting from Google’s imagery, a forgery can inherit contextual cues that make it appear more credible.
Consequences and concerns
- Loss of trust: Services and investigations that relied on Google Earth for fact-checking may now face greater uncertainty about whether imagery has been manipulated.
- Lower barrier to fabrication: Hiding a real facility once required state-level effort; creating a plausible fake one could be done with a single sentence.
- Rapid rollback, lasting lesson: Although Google pulled the feature within 24 hours, the demonstration exposed a structural vulnerability when generative AI is able to edit or augment real-world imagery in place.
Summary
Google briefly allowed an image-generation model to alter satellite imagery in Google Earth via text prompts. OSINT researcher Henk van Ess demonstrated the risk by creating fabricated installations and events with simple prompts; Google removed the capability within a day. The incident highlights how generative AI can undermine longstanding assumptions about the trustworthiness of satellite-based visual evidence.
