Google has withdrawn a new artificial intelligence feature from Google Earth after researchers demonstrated how it could be used to create convincing fake satellite images of real locations.
The tool allowed users to select a place in Google Earth and enter a text prompt describing how they wanted the location to appear. Google’s Nano Banana 2 image model then generated a new visual using the platform’s existing satellite, aerial and three-dimensional mapping data.
Google removed the feature less than a day after its introduction, following criticism from misinformation researchers and open-source intelligence specialists.
The company said some geospatial professionals had found useful applications for the technology. However, it also acknowledged that people were sharing generated images that appeared to violate its policies.
Researchers testing the feature found they could produce realistic images showing events that had never happened.
Examples shared online included fabricated refugee camps near the US-Mexico border, a fictional nuclear facility in Iran and false scenes of destruction in conflict zones.
Because the images were created on top of genuine Google Earth data, they appeared to be connected to real geographic coordinates. This made them potentially more convincing than ordinary AI-generated pictures.
The technology could allow someone to create a fake image of a military attack, natural disaster, protest or humanitarian crisis within seconds. That image could then be removed from its original context and presented online as authentic satellite evidence.
Google said the generated pictures did not replace imagery in the main Google Earth service and were not visible to other users browsing the platform.
However, critics argued that screenshots could still be downloaded and shared across social media, where viewers might not know they had been artificially created.
The controversy is particularly significant because Google Earth has traditionally been treated as a reliable source of visual information.
Journalists, researchers, investigators and members of the public use the service to examine buildings, roads, military sites and environmental changes. Satellite imagery can also help verify reports from areas where access is limited.
Experts warned that placing an image generator inside such a trusted platform could weaken public confidence in genuine satellite evidence.
A fake image created through an unrelated AI application may already raise suspicion. A similar image carrying the appearance and geographic context of Google Earth could be accepted more easily as real.
This could create additional challenges during wars, elections, natural disasters and other fast-moving events when false information spreads quickly.
Google initially defended the feature by pointing to its safety systems.
Images created with the tool included SynthID, Google’s invisible digital watermark for identifying AI-generated content. Users who questioned an image could check it using the Gemini app or Google Lens.
The company also said safeguards were intended to prevent the creation of material involving harmful subjects.
Researchers nevertheless argued that these protections placed too much responsibility on the public. Many users do not verify images before sharing them, particularly when a post supports an existing belief or appears to show a breaking news event.
There were also concerns that watermarks and detection systems may not always remain effective after an image has been cropped, compressed, edited or recorded as part of a video.
One researcher said he was able to produce misleading scenes despite the restrictions and later demonstrated that an external AI detection service did not consistently identify the resulting material.
Google said it was rolling back the feature while it develops stronger protections.
The company has not confirmed when the image generator might return or what additional restrictions would be introduced.
Possible safeguards could include blocking prompts involving conflict, weapons, disasters, borders and sensitive infrastructure. Google could also add more visible labels directly onto generated images or limit access to verified professional users.
Any future version may need to make a clearer distinction between authentic mapping data and fictional visualisations.
The company will also have to decide whether the creative and professional uses of the tool justify the risks associated with realistic geographic misinformation.
Generative AI has already made it easier to create false photographs, voices and videos. The Google Earth feature showed how the same technology could be applied to satellite-style imagery.
This represents a serious challenge for investigators who use visual evidence to determine what has happened at a particular location.
Satellite photographs have often been considered more difficult to manipulate than ordinary images because they come from specialised providers and are linked to identifiable coordinates and dates.
AI tools may change that assumption by allowing anyone to generate believable overhead views without advanced editing skills.
False satellite images could be used to invent military movements, exaggerate damage, create fake construction projects or claim that buildings and facilities exist where none are present.
Google’s rapid decision to remove the feature highlights the difficulties technology companies face when introducing generative AI into widely trusted products.
A tool designed for creative visualisation can quickly become a source of misinformation when it is connected to real places and real-world data.
The incident also shows that invisible watermarks may not be enough to prevent misuse. Clear labelling, strict prompt restrictions and better public verification tools may be required as generated images become increasingly realistic.
Google Earth remains unchanged for ordinary users, and the artificial images created through the experimental tool were not added to the platform’s public mapping database.
Even so, the brief release demonstrated how easily the credibility of satellite imagery could be exploited. As AI generation tools become faster and more accessible, confirming the original source of geographic images will become increasingly important.
Share your thoughts about this article.
Be the first to post a comment!
