Just one day after unveiling a new capability that allowed users to generate AI-powered satellite images with ease on Google Earth, Google has retracted the feature due to concerns over potential misuse. In a statement released on X, the company acknowledged that while geospatial professionals have found valuable applications for this tool, instances of generated imagery being shared in a manner that breaches their policies have emerged. As a result, Google will be suspending this feature while they reinforce safeguards against malicious use.
When initially announcing this functionality, Google described it as an innovative way to engage users' creativity. Product manager Bryan Horowitz detailed that users could simply zoom in on a location in Google Earth, click 'create image,' and input their desired visuals.
However, the ease of creating deepfake satellite images has alarmed experts who utilize Google’s aerial imagery to corroborate urgent news stories and document atrocities in inaccessible areas. Henk van Ess, an open-source researcher who highlighted the potential risks tied to this feature, reported successfully generating images of areas such as the Mexican border and a bombed hospital in Gaza—scenarios that have significant implications if misrepresented.
In tests conducted by NPR, images of non-existent disasters, such as a burning Kharg Island in Iran and a flooded U.S. Capitol, were produced effortlessly. The ramifications of such capabilities raised substantial questions among journalists and open-source investigators alike. Evan Hill, a visual forensics expert at the Washington Post, expressed curiosity about whether the internal testing had thoroughly assessed the risks of abuse and misinformation.
Amid the surge of AI-generated content online, authentic satellite imagery has served as a critical resource for discerning fact from fiction. Open-source investigators heavily depend on these images for validation, often using them to contextualize videos and other information shared online. Though Google Earth does not consistently feature the most current imagery, its high-resolution snapshots can offer vital references for evaluating on-ground events.
Jake Godin, a senior researcher with Bellingcat, emphasized that satellite imagery has historically been a reliable means of event verification due to its difficulty to fabricate. He observed, however, that incidents of fake satellite imagery have already surfaced, particularly in the context of the U.S.-Israel conflict, where manipulated images of a U.S. base in Bahrain circulated—albeit later confirmed to show actual damage, albeit different from the forgeries.
The advent of AI-generated satellite images with just a click poses a substantial risk of amplifying the production of false visuals. Godin voiced his concerns about the potential for confusion and misinformation, noting that erroneous information tends to disseminate quicker and farther than corrections can keep pace with. This may lead governments to dismiss genuine satellite data as fraudulent, further eroding public trust.
In their recent update, Google asserted its commitment to combating misinformation, noting that all AI-generated images will carry a SynthID watermark to indicate their artificial origin. The company also claimed to prevent the creation of images related to harmful subjects and promised ongoing updates to their protective measures.
Despite these assurances, many in the open-source community remain unconvinced. Van Ess highlighted the significance of Google Earth as a trusted verification tool for assessing the authenticity of information shared on social media. He cautioned that if Google continues to advance this AI capability, it could undermine the very platform’s credibility that experts rely upon to discern fact from fiction.
