Google pauses AI satellite images, after fears of deepfakes in the sky : NPR

Just one day after Google introduced a feature on its Google Earth platform that enabled users to generate AI-created satellite images with a simple click, the company decided to withdraw the tool.

“We’ve observed geospatial professionals utilizing this feature for a variety of beneficial purposes,” Google stated in a statement on X. “However, we have also seen individuals sharing screenshots of generated imagery that seem to breach our policies. Consequently, we are retracting this feature in Google Earth while we work on implementing more robust safeguards.”

In its initial announcement on Thursday, the company described the feature as designed to inspire users’ imaginations: “Simply zoom in to a location on Google Earth on the web, click ‘create image,’ and type what you wish to see,” wrote product manager Bryan Horowitz.

However, for experts and analysts who depend on Google’s imagery to confirm breaking news and atrocities in remote areas, the ability to create deepfake satellite images with a click was alarming.

“I tested scenarios like refugees at the Mexican border, a nuclear plant in Iran, an accident in Amsterdam, a hospital with a bomb crater in Gaza. Nothing was turned down,” Henk van Ess, an open-source researcher who first highlighted the potential harm the tool could cause, told NPR via text.

NPR easily generated images depicting Iran’s Kharg Island on fire and a flooded U.S. Capitol complex. Both events, if true, would be significant news.

Online, journalists and open-source investigators questioned Google’s decision. “Very curious how or if this idea was red teamed internally because the opportunities for abuse and disinfo are literally boundless,” Evan Hill, a visual forensics investigator at the Washington Post wrote on X.

Eyes in the sky

As the digital world becomes saturated with AI-generated deepfakes, satellite imagery remains a crucial tool for discerning reality. Open-source investigators use satellite images to verify on-the-ground events and contextualize videos and images shared online.

Although Google Earth imagery may not always be the most current, its comprehensive, high-resolution coverage of the globe offers valuable reference imagery for both ground-level images and more recent satellite images.

“Satellite imagery has been kind of a safe bet when it comes to verifying an event because it’s hard to fake,” says Jake Godin, a senior researcher with the online group Bellingcat, which conducts visual forensic investigations. Satellite images are typically managed by companies or institutions, and their origin—a camera located hundreds of miles above Earth—has traditionally been difficult to replicate.

Fakes from space

Godin noted that fake satellite imagery has already surfaced in certain instances. Early in the U.S.-Israel conflict with Iran, for example, fabricated images spread of a damaged U.S. base in Bahrain. The images, appearing in Iranian media, seemed to use Google Earth images and Google AI to create fictitious damage at the base. (Later, authentic satellite images confirmed the base was damaged, though not as depicted in the fake image.)

An AI-generated image showing significant flooding around the U.S. Capitol.

Generated and annotated by NPR

hide caption

toggle caption

Generated and annotated by NPR

However, the ease of creating AI-generated satellite images with a single click could significantly speed up the production of AI fakes, Godin pointed out. “It’s just streamlining the process, which I think is going to make it proliferate more.”

Godin expressed concerns that the proliferation of fake images would contribute to confusion and misinformation. “Misinformation travels further and faster than any sort of correction,” he noted. Additionally, governments might be more inclined to dismiss satellite images as fake, even if they are authentic, which could erode public trust.

In a post on X on Thursday, the company emphasized its commitment to tackling misinformation. It stated that all images are marked with a SynthID watermark, ensuring they will be recognized as AI-generated in other Google tools like Gemini. “Moreover, we prevent image creation on harmful topics and are continually updating our protections,” it added.

However, many in the open-source community remained skeptical of Google’s assurances. For van Ess, the decision is particularly disappointing because Google Earth has long been considered one of the best tools for verifying information shared on social media.

If Google proceeds with developing the AI tool, van Ess noted, it would mean that “the fake is made inside the thing people use to check whether pictures are true.”