Google Earth’s AI image-editing feature lasted less than one day. Launched on July 30, the tool was already generating convincing fake images of refugee camps and nuclear facilities before Google pulled it the same day. This incident is a concrete test case for a problem the AI industry has not resolved: how to grant creative power over authoritative platforms without enabling fabrication at scale.
What did Google launch, and why was it pulled so fast?
On July 30, Google introduced an AI feature inside Google Earth that let users edit imagery of real locations using plain text descriptions. The tool was powered by a model called Nana Banana 2. Type a description, and the model would render it onto actual satellite or street-level imagery of a real place.
Within hours, AI and open-source data researcher Henk van Ess had used the tool to produce images of what appeared to be refugee concentration camps along the US-Mexico border and a fully operational nuclear power plant at a previously empty site in Iran, according to reporting by Business Insider, Reuters, and Wired. Other social media users generated imagery of floods that never occurred and Godzilla attacking downtown Los Angeles.
- Feature launch: July 30
- Time to forced pullback: less than 24 hours
- Model involved: Nana Banana 2, integrated into Google Earth
- Documented fake types: geopolitical scenes (border camps, nuclear sites), non-existent disasters, fictional events
Google confirmed that AI-generated images carried watermarks (digital labels embedded in the image file to identify AI-generated content) and were not surfaced to regular users within the Google Earth interface itself.
What did Google say?
“We’ve seen experts use the feature for useful purposes, but also discovered many people sharing fake photos and violating company policies,” a Google spokesperson said, as reported by Business Insider, Reuters, and Wired. “We decided to remove the feature on Google Earth while developing stronger safeguards.”
The statement confirms Google anticipated some misuse. What it underestimated was the speed. The gap between “policy-compliant use” and “policy-violating use” closed faster than moderation could respond. Watermarking addressed attribution after the fact; it did not prevent the initial spread.
Is this a pattern?
It is not isolated. On July 7, Meta launched Muse Image, described as the first free image-generation tool from Meta Superintelligence Labs, according to VnExpress citing Business Insider, Reuters, and Wired. One capability allowed users to edit photos from other public Instagram accounts simply by tagging the account. Muse would pull the public image and generate an AI-altered version.
Privacy concerns emerged almost immediately. Meta said users could adjust settings to block others from generating AI images from their posts without making the account fully private. Three days later, the feature was locked entirely.
Two major AI image tools, two forced rollbacks within weeks of each other. The pattern suggests that product teams are consistently underestimating how quickly intended use diverges from actual behavior when the feature is released at scale to a general audience.
What this means for AI-search visibility
The core issue is one of authority: mapping platforms carry a level of credibility that social media does not, and AI systems often treat data from authoritative platforms as more reliable.
Think of it this way. If a fake image of a refugee camp is posted on X, most readers apply at least some skepticism. If the same image appears to originate from Google Earth, that skepticism drops sharply, because maps and satellite tools carry an implicit “this is what is actually there” assumption. That trust transfer happens automatically, without verification.
This is a Hingewise observation, not stated in the source reporting: the risk extends beyond Google’s reputation. Geographic and satellite imagery are increasingly used as grounding data (reference information that AI systems treat as factual anchors when answering location-based queries). If manipulated imagery from a trusted mapping source enters that layer, even temporarily, the false information can persist in AI-generated answers long after the original fake image is removed from the platform. AI systems may have indexed or cached that content before the pullback.
For businesses with physical locations, the implication is practical: how your address and premises are visually represented across authoritative platforms, including Google Maps, Google Earth, and similar tools, is now part of your AI-search footprint. If fabricated imagery tied to your real address circulates and gets indexed, it can surface in AI responses to location queries, independent of anything you control on your own website.
The broader structural issue, in Hingewise’s assessment, is that the “launch, observe, patch” approach is not appropriate for platforms that carry map-level authority. Social media platforms learned this lesson over years of content moderation challenges. Mapping and geographic platforms are now entering the same cycle, with higher baseline trust and, potentially, higher consequences when that trust is exploited. Expect pre-launch safety testing requirements to tighten for AI features on location platforms, and this category of incident to draw closer regulatory attention as examples accumulate.
Before your brand’s location shows up in an AI-generated context: a quick checklist
- Query your brand name and physical address in major AI tools (ChatGPT, Gemini, Perplexity) and note how your location is currently described or depicted.
- Verify that your Google Business Profile and official website carry accurate, up-to-date visual and written location information, as these are common retrieval sources for AI systems.
- Set up a Google Alert combining your brand name with terms like “fake image,” “deepfake,” or your street address to catch early signals of fabricated content.
- Audit whether your public imagery on Instagram, Google Maps, or similar platforms is sourced from your own accounts and accurately represents your premises.
- If you discover fabricated imagery linked to your location, document it with timestamps and report it directly to the platform before requesting removal.
- Check with your communications or legal team on whether AI-generated location misinformation falls under existing brand protection, defamation, or false advertising policy in your market.
Google’s Google Earth AI feature was live for less than 24 hours. The fabricated images it enabled are still in circulation. That gap between how long a feature lives and how long its content lives is the variable worth tracking as more mapping and location platforms add AI-editing capabilities.
Sources: VnExpress (July 2026), citing Business Insider, Reuters, and Wired. Original reporting: VnExpress.
Lam Nguyen · Hingewise