Google announced it was removing the AI image generation feature from Google Earth one day after its launch. The company stated it is rolling back the feature while working on implementing stronger guardrails.

Google introduced an AI image generation feature into the web version of Google Earth on Thursday. The feature allowed users to type a text prompt to generate new images over existing satellite, aerial, and 3D imagery in Google Earth. According to internal documentation, the tool was explicitly labeled as experimental in the web interface.

The capability relied on Google's Nano Banana 2 model to produce visual overlays. Users could manipulate the virtual map program developed by Google to insert synthetic elements into real-world geographic contexts. It stated that generated images did not appear in the main Google Earth experience for others to see.

Henk van Ess, an open-source researcher, generated an image showing a bomb crater near a hospital in Gaza using the same feature. He reported that the system accepted prompts for sensitive geopolitical scenarios without restriction. "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," van Ess said in a text message.

He added that nothing was softened and nothing suggested he try a different prompt during his testing. Ben Heubl, an OSINT professional, noted the potential for localized disinformation. "If you type the address of your workplace and then prompt it to ‘make it look like a drone attack happened here,’ it will create a damning image that looks very close to what we’ve seen in Ukraine, Russia, the Middle East, and other war-torn conflict zones," Heubl said in a LinkedIn post. He noted that such images could closely resemble footage from actual conflict zones.

It had implemented safeguards intended to identify synthetic media. The company stated that every image created with the feature includes a SynthID digital watermark. "We take misinformation seriously – every image created with Nano Banana in Google Earth includes the SynthID digital watermark, so if someone is unsure about an image, they can ask the Gemini app or use Lens in Search to see if the image was AI-generated. In addition, we prevent image creation on harmful topics and are continually updating our protections," it said on social media platform X.

Critics questioned the reliability of these detection measures. Henk van Ess stated that he was able to fool Hive’s AI detection tool with a video generated in Google Earth. Ben Heubl said there are ways to override AI detection indicators.

"Even if you can use detection tools to determine that an image was altered by Google, there are ways to override those indicators, no matter what people tell you. This is BAD," he said.

The limitations of the watermarking system raised further concerns. Internal documentation shared with Ars Technica indicated that Google's SynthID system limited users to 10 image verification checks per day during testing. Reports suggest the system covers less than 25% of AI-modified images in tested examples, which limits full verification. In May 2026, Google's SynthID AI watermark standard was adopted by OpenAI, Kakao, and ElevenLabs to improve content authentication across platforms.

It explained the decision to remove the tool by citing policy violations. "We've seen geospatial professionals using this feature for a range of useful purposes. However we've also seen people sharing screenshots of generated imagery that appear to violate our policies. So we're rolling back this feature in Google Earth while we work on implementing stronger guardrails," it said.

Why It Matters

The rapid removal of the feature shows the difficulty of preventing AI tools from generating realistic images of sensitive geopolitical events, such as attacks on hospitals or border crises. Experts note that while Google implemented digital watermarks, these indicators can be overridden or fooled, limiting their effectiveness against disinformation. This incident demonstrates how easily synthetic media resembling actual conflict zone footage can be created and shared before safeguards are fully operational.