A prompt-based tool that generated fake satellite imagery inside Google Earth was rolled back within about a day, after analysts warned it let anyone fabricate warzones, disasters, and facilities on a map people trust to be real.
2026-07-31 · by Moe Ameen
Google added an AI image-generation feature to the web version of Google Earth, letting users type a prompt and overlay fabricated satellite imagery onto real map coordinates. The tool was powered by Nano Banana 2, Google's image model, and could make a neighborhood appear bombed, drop a facility into a location that has none, or paint a city as the site of a natural disaster. It launched on or around Thursday, July 30, 2026, and Google rolled it back roughly a day later.
The reversal followed fast, public criticism from open-source intelligence (OSINT) researchers and journalists. Google Earth's satellite view is one of the tools reporters and analysts use to verify claims from conflict zones and disaster areas, so a generator that fabricates imagery inside that same interface undercuts the thing the map is trusted for. OSINT researcher Henk van Ess demonstrated the risk by generating fake refugees near the US–Mexico border, placing a facility in Iran, and adding a bombed hospital to imagery of Gaza — noting the output "inherits the credibility of the map it was born on."
Google said in a statement that it had "seen geospatial professionals using this feature for a range of useful purposes," but also "seen people sharing screenshots of generated imagery that appear to violate our policies," and was "rolling back this feature in Google Earth while we work on implementing stronger guardrails." The company added that people "uniquely trust Google Earth for a reliable view of the world."
Google noted that AI-generated images carried its SynthID provenance watermark, in principle detectable via Gemini or Google Lens. In practice, reporting on the tool found the checks unreliable — Gemini sometimes returned that it was unsure whether an image was AI-made or found "no reliable signals" — which is part of why watermarking alone did not settle the concern. Exact launch timing and the feature's return date should be confirmed against Google's own posts, since the rollout and rollback moved within a single news cycle.
The lesson here is not "avoid AI images" — it is "don't hang your workflow on one experimental button that can vanish in a day, and own the labeling and distribution yourself." Kompozy is a full AI content generation and multi-platform publishing engine, so the image step sits inside a durable pipeline rather than a single provider feature. It generates Photo Posts, Infographic Photos, Quote Graphics, brand-exact Carousel Posts, and face-locked Persona Photos through its own model stack, then reframes each to 9:16, 1:1, and 16:9 for where it's going.
The provenance and disclosure problem is exactly where the pipeline matters. Every asset runs through a per-post review pipeline before it publishes, so you caption, add an AI-generated disclosure, and check the framing in one place instead of trusting a viewer to reverse-check a watermark. From there Autopilot schedules and publishes across nine destinations — the eight primary social platforms plus blog and email — on a cadence you set. When a headline tool gets pulled overnight, a creator on Kompozy loses nothing: the generation, labeling, and distribution all live in a workflow they control, not in a feature that shipped and disappeared in the same news cycle.
It was a prompt-based tool in the web version of Google Earth that generated fake satellite imagery and overlaid it on real map coordinates, powered by Google's Nano Banana 2 model. Google rolled it back about a day after launch after OSINT researchers and journalists warned it could spread misinformation on a platform people trust for a reliable view of the world.
Google described it as a rollback while it works on "stronger guardrails," not a permanent removal, which suggests a possible return in a more constrained form. Because the launch and rollback happened within a single news cycle, confirm the current status against Google's own announcements before relying on it.
Yes — Google said generated images carried its SynthID provenance watermark, detectable in principle via Gemini or Google Lens. But most viewers never run such a check, and reporting on the tool found the checks themselves unreliable, so watermarking alone did not resolve the misinformation concern.