Google's Nano Banana AI Generator in Earth Creates Misinformation Risk Despite Watermarks

Nano Banana 2 Arrives with Misinformation Potential

Google launched Nano Banana 2, an AI image generator embedded directly in Google Earth, allowing users to synthesize satellite and aerial imagery from natural-language text prompts. According to The Verge, the feature went live globally on the web version of Google Earth in late July 2026. Google positions the tool for visualizing historical sites and real estate projects, but security researchers have quickly demonstrated its capacity to generate convincing false imagery—including fabricated scenes of “refugees near the Mexican border” and bomb craters near hospitals, as documented by Digital Digging’s Henk van Ess.

Watermarks and Detection Gaps

Google has embedded SynthID digital watermarks into every image created with Nano Banana, framing the watermark as a disclosure mechanism for downstream verification. The company recommends users employ its @verifyai tag in Gemini or Google Lens in Search to identify synthetic content. According to The Verge, Google stated in a company statement: “We take misinformation seriously – every image created with Nano Banana in Google Earth includes the SynthID digital watermark, so if someone is unsure about an image, they can ask the Gemini app or use Lens in Search to see if the image was AI-generated.”

However, the technical defenses show cracks. Digital Digging researchers successfully bypassed Hive’s AI detector using a video derived from Nano Banana imagery, suggesting watermarks can be removed or overlooked. The Verge notes that van Ess recommends a multi-layered verification approach: cross-referencing outputs against satellite platforms like Sentinel-2 or Landsat and validating orbital metadata—timestamps and satellite identifiers embedded in purported images.

The Believability Problem

The core risk is not that all Nano Banana outputs look real, but that some do. Extreme forgeries—such as the Sphinx merged with the Statue of Liberty—are obvious fakes. Yet intermediate cases, particularly those blending plausible geographic scenes with subtle alterations, can escape casual scrutiny and circulate as documentation of real events before fact-checkers intervene. The Verge highlights this distinction: the tool’s integration into an authoritative source (Google Earth) and its reliance on real satellite geographies make synthetic outputs inherently more credible than purely fictional generative images.

Why This Matters

News organizations, humanitarian groups, and verification platforms now face a new verification burden. While SynthID watermarks are difficult to fully remove, they are not impossible to overlook or bypass—a gap that content moderation and fact-checking infrastructure must address through redundant verification layers. Teams relying on satellite imagery for evidence (investigative journalists, conflict monitors, real estate analysts) will need to adopt the multi-step verification practices Henk van Ess has outlined. The broader implication is that trustworthiness of satellite data cannot rest solely on platform-side watermarking; it requires distributed verification ecosystems and user literacy about synthetic-media detection.