Google has added an AI image generator directly into Google Earth, the same tool journalists and investigators have used for years to figure out what's actually real. With a single click, anyone can now produce a convincing fake satellite image of, say, Iran's oil infrastructure on fire, or the U.S. Capitol submerged in floodwater. Neither of those things has happened. Both images now exist.
What Google Actually Built Here
The feature launched Thursday. According to NPR, the announcement from Google product manager Bryan Horowitz described it with the chirpy simplicity of a children's app: "Just zoom in to a place in Google Earth on web, tap 'create image,' and type whatever you want to see." Zoom in. Click. Type. Done. You now have a satellite photograph of something that never happened.
NPR tested it and generated images of Kharg Island, Iran's critical oil terminal in the Persian Gulf, engulfed in flames. They also produced a flooded U.S. Capitol. Both are fabrications. Both look, at a glance, like exactly the kind of imagery that gets screenshotted, stripped of context, and fired across social media the second a real crisis breaks.
Open-source researcher Henk van Ess, who was among the first to flag the problem publicly, ran his own tests. He told NPR via text: "I tried refugees at the Mexican border, a nuclear plant in Iran, a crash in Amsterdam, a hospital with a bomb crater in Gaza. Nothing was refused." Nothing. Refused.
Why Satellite Imagery Was the Last Safe Ground
Here is the thing about satellite images: they were supposed to be the reality check. In a media environment drowning in AI slop and doctored photos, verified satellite imagery from Google Earth has been one of the few tools investigators could point to and say, with confidence, this is what is actually there.
Jake Godin, a senior researcher at Bellingcat, the open-source investigative group that has used satellite imagery to document everything from war crimes to missile strike locations, explained it plainly to NPR. "Satellite imagery has been kind of a safe bet when it comes to verifying an event because it's hard to fake." The camera is hundreds of miles up in space. The data runs through institutions and companies with paper trails. Historically, that has made it extremely difficult to convincingly fabricate.
And yet. Godin told NPR that fake satellite images have already started appearing in limited cases. Early in the U.S.-Israel conflict with Iran, fabricated images purporting to show damage at a U.S. base in Bahrain circulated in Iranian media, apparently using Google Earth imagery and AI-generated damage layered on top. Real satellite images later confirmed the base was in fact struck, just not in the way the fakes depicted. So the fakes existed alongside the truth, and the truth had to chase them down.
One Click to Flood the Zone
What Google has done here, whether it thought this through or not, is compress the work. Creating a convincing AI-generated fake used to require some technical effort, some time, maybe some skill. Google has automated that entirely. Godin put it simply to NPR: "It's just streamlining the process, which I think is going to make it proliferate more."
Proliferate is the polite word. What he is describing is a world where the barrier to manufacturing a fake satellite image of a hospital bombing, a naval fleet, a refugee camp, or a nuclear facility is now functionally zero. You need a Google account and a creative impulse.
Evan Hill, a visual forensics investigator at the Washington Post, wrote on X that he was "very curious how or if this idea was red teamed internally because the opportunities for abuse and disinfo are literally boundless." That is a professional, careful person using the word "boundless." That should alarm you.
Google's Answer, Such As It Is
Google did not respond to NPR's request for comment before publication. The company did post on X that it "takes misinformation seriously," and noted that images generated by the tool are stamped with its SynthID watermark, which means other Google products like Gemini will flag them as AI-generated.
That is the whole defense. A watermark. On a platform where images get downloaded, cropped, screenshotted, reposted, and stripped of metadata approximately one thousand times before breakfast. The SynthID watermark is invisible to the human eye and depends on other Google tools being present in the chain to detect it. It will not stop a bad actor. It will not stop a credulous re-poster. It will not stop a state media outlet that wants a fake satellite image of a burning oil terminal right now, tonight, during a fast-moving military conflict.
Google also told NPR it "prevents image creation on harmful topics" and is "continually updating" its protections. Van Ess's test results, in which he generated images of a bombed Gaza hospital and an Iranian nuclear plant without any refusals, suggest those protections are not exactly airtight at launch.
The Trust Collapse Is the Real Weapon
Godin raised a second, uglier problem to NPR that goes beyond the fakes themselves. Once AI-generated fake satellite images become common knowledge, governments and officials gain a new and powerful rhetorical weapon: claiming that real satellite images are fake.
This is not a hypothetical. Authoritarian governments have been dismissing inconvenient photographic evidence for years. Now they have a plausible technical argument. "Sure, that image looks like our forces destroyed a village. Could be AI. Google does that now." The doubt is the point. The confusion is the product.
"Misinformation travels further and faster than any sort of correction," Godin told NPR. That was already true before this week. Google has now handed misinformation a vehicle with a turbo engine and no speed limiter, inside the very tool people used to slow it down.
The Dingo Take
Van Ess said it best to NPR, and it deserves to be repeated until someone at Google has to read it out loud: "The fake is made inside the thing people use to check whether pictures are true." That is not a bug report. That is an epitaph for one of the last credible verification tools the internet had left.
Google is a company that employs some of the smartest engineers on earth, runs one of the most sophisticated AI safety research organizations in the industry, and apparently did not think hard enough about what happens when you let anyone generate a fake satellite image of a nuclear facility with zero friction during an active period of Middle East conflict. The SynthID watermark defense is insulting. It assumes a world where misinformation moves slowly and politely through Google-approved channels. That world does not exist and has not existed for about a decade.
Somebody greenlit this feature. Somebody looked at the product spec, saw "users can type whatever they want to see" applied to satellite imagery of real places during real geopolitical crises, and signed off. That person should be having a very uncomfortable Thursday. Instead, Google posted a cheerful blog entry and waited to see if anyone noticed. We noticed.
Comments