On 30 July 2026, Google switched on Nano Banana in Google Earth Web, a tool that overlays AI-generated scenes on real satellite coordinates. The next day, the tool was withdrawn after fabricated scenes deemed contrary to the platform's policies spread online, a setback that calls into question the value of any image anchored to a real place.
The facts
- Google launched Nano Banana in Google Earth Web on 30 July 2026, a tool generating photorealistic images on real geographic coordinates.
- In under 24 hours, media outlets and researchers (404 Media, NPR, BBC Verify) produced fictional scenes: a crater in Los Angeles, a flooded Capitol, a collapsed Eiffel Tower.
- Google withdrew the tool on 31 July 2026 and announced it would only bring it back after strengthening the guardrails, with no timetable given (Newsweek).
Why does anchoring to a real place change the nature of the risk?
An image generated on a neutral background remains identifiable as such. Placed on a recognised satellite map that has served for years as a trusted reference, it changes status. Researcher Henk van Ess, who tested the tool before its withdrawal, sums up the mechanism: a fabrication has no need to be convincing in itself, since it borrows the credibility of the medium it is built on. That transfer of trust, more than the technical quality of the rendering, is what alarmed disinformation researchers within the first hours.
Google had built in a SynthID watermark meant to flag any generated content, verifiable via Gemini or Google Lens. Tests by BBC Verify showed it could be circumvented in some cases, and that external detection tools also missed some fabricated images. Researcher Henry Ajder notes that in a crisis, few people take the time to verify an image before sharing it. The watermark protects the source that knows where to look, and does little for the public that discovers the image in its feed.
AIxH's view
This episode shifts the problem from the technical field to process, namely who validates an AI-generated image before it is published. That is what our AI integration in business service helps to frame, since a watermark is no substitute for human verification. The question deserves to be asked before an incident asks it for you.