Google Earth's AI Image Tool Lasted 24 Hours — Here's What It Did to Singapore's Skyline

For two decades, satellite imagery has been one of the last bastions of visual truth on the internet. When an open-source investigator needs to verify whether a building was destroyed in a conflict zone, they pull up Google Earth. When a journalist needs to corroborate a claim about a location, they reach for satellite data. That trust — painstakingly built over 20 years — was put at risk on Thursday, when Google quietly rolled out an AI feature that let anyone generate fake satellite images with a single text prompt. By Friday, the feature was gone. But for one day, anyone could have flooded social media with scenes of destruction at Singapore's most recognisable landmarks.

Google integrated its Nano Banana AI image generator directly into Google Earth, allowing users to type prompts like "fire at Gardens by the Bay" or "collapsed bridge at Marina Bay Sands" and receive photorealistic satellite-style images in seconds. CNA tested the tool before it was pulled and generated three fictitious disaster scenes at Singapore locations without encountering a single refusal or restriction. The publication then ran the screenshots through four commercial AI-detection tools: half deemed two of the images as authentic with low likelihood of manipulation, and one flagged an image as likely real. Google's own Gemini app did detect the AI origin, but by then the screenshots could already be uploaded to Facebook, Instagram, or TikTok without any requirement to label them as AI-generated.

The episode highlights a growing asymmetry in the AI disinformation landscape. For years, satellite imagery has been the tool used to debunk falsehoods — investigators in Ukraine and the Middle East have relied on Google Earth to verify claims about troop movements, destroyed infrastructure, and mass graves. When Iran's state-run media posted a fake Google Earth image in February claiming a drone strike had destroyed US radar equipment in Qatar, it was the credibility of genuine satellite data that made the fabrication convincing. Embedding an AI-generation tool directly into the same interface doesn't just lower the barrier to creating fakes; it erodes the very platform people turn to for verification.

Benjamin Ang, who heads the centre of excellence for national security at Singapore's RSIS think-tank, told CNA that satellite images have always carried an "objective angle" that made them harder to tamper with — and therefore more believable. "In the past, making fake satellite images in a convincing way needed a relatively high level of resources, technology, and skill," he said. "Now practically anyone can create fakes that cause panic or outrage." The timing is especially concerning for Singapore, which has been strengthening its own defences against online falsehoods through POFMA and public media literacy campaigns. A tool that weaponises the very format citizens are taught to trust represents a new category of threat — one that legislation written for social media posts may struggle to address.

Why it matters for Singapore: The Google Earth incident is a preview of how AI image generation will challenge Singapore's information integrity infrastructure. The government's $50 million Centre for Advanced Technologies in Online Safety (CATOS), launched in 2024, will need to contend with AI-generated satellite imagery alongside deepfakes and text-based disinformation. And as Singapore positions itself as a trusted digital hub — a recurring theme in recent IMDA and MCI policy announcements — incidents that undermine trust in widely-used platforms like Google Earth have direct implications for the city-state's credibility as a digital gateway for the region.