Alphabet pulls Earth AI image tool after 24 hours as users fabricate warzone and disaster scenes

The feature and its launch

Alphabet's Google introduced an AI image generation tool inside Google Earth on Thursday, July 30, 2026. The feature, powered by the company's Nano Banana 2 AI model, allowed users to zoom into any location on the globe, type a text prompt, and generate a photorealistic aerial scene grounded in Google Earth's satellite, aerial, and 3D imagery. Bryan Horowitz, product manager for Google Earth, described it in a July 30 company blog post as a way to 'generate custom images using Google Earth's satellite, aerial, and 3D imagery alongside Nano Banana, which creates concepts grounded in the real world.'

Google initially presented the tool as useful for geospatial professionals, real-estate visualization, and educational scenarios such as reconstructing Pompeii in 78 CE. The generated images did not appear in the main Google Earth base layer and carried Google's SynthID digital watermark, an invisible signature designed to survive screenshots and compression.

Abuses surface within hours

Within a day, independent researchers and news organizations demonstrated how single-sentence prompts could produce alarming and deceptive imagery. Dutch intelligence researcher Henk van Ess published a blog post titled 'How to plant a nuclear plant in Iran,' showing prompts that inserted a nuclear facility into Iranian territory, placed refugees on a road near the Mexican border, and added a hospital with a bomb crater in Gaza, each in seconds.

BBC Verify generated a toppled Eiffel Tower, a sinkhole swallowing the Great Pyramid of Giza, and Russian tanks in Kyiv. AFP produced explosions in Paris and an Islamic State training camp in Syria. 404 Media inserted a homeless encampment into Los Angeles and skyscrapers onto rural land. The Washington Post reported users creating what looked like a plane crashing into a Manhattan skyscraper and aerial images of a bombing in Moscow. In every case, the fabricated imagery was layered directly onto authentic coordinates.

Google spent twenty years building the reference the whole world trusts. Now it is adding a button that invents things.

The concern was not the AI's ability to generate fake images, which has existed for years, but the welding of fabricated scenes onto real coordinates inside a platform that journalists, investigators, and intelligence analysts treat as a primary verification tool. Van Ess noted that it takes a government request to remove a single real building from Google Earth, but only a sentence to add a fake one.

Google's response and rollback

Google initially did not remove or restrict access to the tool. The company stated on its media relations account on X that it takes 'misinformation seriously' and directed users to check suspicious images using the Gemini app or Lens in Search to detect the SynthID watermark. It also said it blocked 'creation of imagery on dangerous topics' and pointed to an AI use policy that prohibits disinformation and misleading activities.

By Friday, July 31, after The Atlantic reached out for comment and as fabricated images circulated widely, Google reversed course. The company posted on X: '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.' Google added it was 'rolling back this feature in Google Earth while we work on implementing stronger guardrails.' Shortly afterward, Ars Technica confirmed the feature was inaccessible to users.

The proliferation of these AI-generated fakes could also erode public trust in satellite imagery and complicate the work of journalists and researchers.

Brady Africk, a satellite imagery analyst at the American Enterprise Institute, warned AFP that the update would have made it easy to 'create convincing fake satellite images that could spread rapidly online and deceive the public.' Eliot Higgins, founder of Bellingcat, posted a sarcastic Bluesky response to the debut that featured an AI-modified image showing a giant golden statue of President Donald Trump over the White House, writing: 'No way this could be abused.'

Credibility and verification at risk

Google Earth has been a unique resource for researchers, journalists, and intelligence officials studying armed conflict and natural disasters. Satellite and aerial imagery is used to monitor crop failure, deforestation, mass graves, and the results of bombing campaigns, especially in areas that are hard to reach, where satellite imagery often acts as the closest available ground truth. The integration of an AI image generator that could layer fabricated scenes onto verified coordinates threatened that function, with one investigative reporter writing on LinkedIn that the tool could rapidly create images of drone strikes.

The incident follows a similar trajectory at Meta, which earlier in July discontinued its Muse Image AI feature after privacy concerns and criticism from a Hollywood union. Across the tech sector, companies have rushed to integrate generative AI into consumer products and have repeatedly had to adjust, restrict, or disable features in response to user violations and misinformation risks.

Google has not specified what types of images triggered the policy violations or when the feature might return. The company said the AI-generated images were confined to the generation tool and never mixed into the main Google Earth base imagery.