Google launched a Google Earth web feature on July 30 that lets users generate images from satellite, aerial, and 3D map views using a text prompt. NPR, AFP, and open-source researchers documented prompts that produced false depictions of fires, crashes, flooding, and other sensitive events. Google cited SynthID watermarking, Gemini and Lens detection, and content guardrails as safeguards.
Google has launched a web-based Google Earth feature that lets users generate custom images from a selected location's satellite, aerial, and 3D imagery using a text prompt. In its July 30 product announcement, Google presented the capability as a way to visualize historical scenes, create location-based infographics, and develop real-estate concepts.
The feature combines Google Earth views with Google's Nano Banana image-generation technology. Google instructs users to zoom to a location, select "create image," and describe what they want to see. Its examples include recreating Pompeii in 78 A.D. and reimagining a concrete area as a retail district.
Researchers demonstrate misleading scenes
The release immediately prompted concern from journalists and open-source intelligence researchers, who use satellite imagery to corroborate events in places that are difficult to access. NPR reported that it generated images depicting fires on and near Iran's Kharg Island and flooding at the US Capitol complex, neither of which had occurred.
Henk van Ess, an online-research specialist, wrote that he used prompts to create depictions of refugees near the Mexican border, a nuclear plant in Iran, a fatal crash in Amsterdam, and a bomb crater near a Gaza hospital. He wrote, "Google spent 20 years building the reference the world checks against. Today it added a button that makes things up."
AFP reported that its own tests generated fabricated satellite images with potentially serious geopolitical implications. Brady Africk, a satellite-imagery research analyst at the American Enterprise Institute, told AFP that the update "makes it easier to generate convincing fake satellite imagery that can spread quickly online and mislead the public."
The concern is not that Google Earth has altered its underlying imagery archive for every user. Rather, the reporting describes a new workflow that can produce a plausible synthetic scene while retaining the visual cues, framing, and geographic context associated with a widely used mapping product.
Google's stated safeguards
Google responded to criticism by citing its provenance and detection controls. According to statements reported by The Verge, AFP, and The National, every image created with Nano Banana in Google Earth includes a SynthID digital watermark. Google said users who are unsure about an image can ask the Gemini app or use Lens in Search to determine whether it was AI-generated.
Google also stated that it blocks image creation on harmful topics and that its AI policy prohibits misinformation, misrepresentation, and misleading activity. The National reported that Google described its guardrails as continually updated.
Those controls have practical limits in a fast-moving verification environment. SynthID is designed to be machine-detectable, not necessarily obvious to a person viewing a reposted image. In comparable misinformation incidents, provenance checks require the recipient to know that a check is needed, have access to a compatible tool, and retain enough of the original asset for detection to work.
A verification challenge for geospatial workflows
For OSINT teams, newsrooms, and incident-response analysts, the relevant question is whether a satellite-like image is authentic source imagery, a visualization, or an AI-generated artifact. Van Ess told NPR that prompts involving a nuclear plant in Iran, refugees at the Mexican border, a crash in Amsterdam, and a Gaza hospital bomb crater were not refused in his testing. NPR reported that Washington Post visual-forensics investigator Evan Hill questioned whether the feature had been red-teamed internally, calling the opportunities for abuse and disinformation "literally boundless." That statement reflects an external expert's assessment, not a disclosed account of Google's internal testing.
Practitioners handling geospatial evidence already compare imagery across sources, inspect acquisition dates and satellite metadata, and validate orbital or sensor context. Van Ess recommended comparisons with other satellite platforms and checks of when an image was captured and by which satellite. As image-generation systems become embedded in mapping interfaces, those established practices become more important because visual realism alone is not evidence of an event.
Key Points #
- 1Google Earth now generates prompt-driven location images, creating a new path for synthetic scenes to borrow mapping context and visual credibility.
- 2NPR, AFP, and researcher tests documented fabricated crisis imagery, showing that safeguards did not prevent every sensitive prompt tested.
- 3Geospatial verification teams increasingly need provenance checks, cross-platform comparisons, and acquisition metadata because realistic imagery alone cannot establish authenticity.
Scoring Rationale #
Embedding generative imagery in a major mapping platform creates a material verification and misinformation risk for OSINT, journalism, and geospatial analysis workflows. The feature is globally available on Google Earth web, while multiple sources documented sensitive fabricated outputs shortly after launch.
Sources #
Primary source and supporting public references used for this report.
View 5 more sources #
Google adds AI to satellite images, raising fears of deepfakes in the skynpr.orgHere’s the problem with putting an AI image generator in Google Earththeverge.comHow to plant a nuclear plant in Irandigitaldigging.orgGoogle launches new satellite image AI tool, alarming ...france24.comGoogle Earth AI controversy: Company responds to misuse ...thenationalnews.com
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.