Can we trust Google Earth as evidence anymore?

In partnership with

Last week, Google added a new feature to Google Earth. "Generation of custom images grounded in Google Earth's satellite, aerial and 3D imagery" with its image generation tool Nano Banana.

According to Bryan Horowitz, a Product Manager at Google Earth and My Maps, "the experience is simple."

Just zoom to a place in Google Earth, tap 'Create Image', and type whatever you want to add. Horowitz seemed delighted about what people would come up with, and signed off with a hashtag: #goBananas.

That feature has since been removed, following major backlash. Because it turns out you could easily create nuclear power plants, bomb craters, large numbers of people crossing border regions, and all sorts of stuff that would turn an already strained information ecosystem into a disinformation free-for-all.

So what exactly shipped? You zoomed to any coordinates on Google Earth's web version, clicked 'Create Image', and typed a prompt. Nano Banana 2, Google's newest image model, took Google's real satellite, aerial and 3D imagery of that exact spot as its base layer and handed back a photorealistic picture…. same light, same angle, same colours as the genuine scene beside it.

Google called these "concepts grounded in the real world". Suggested uses included Pompeii in 78 A.D., infographics and real-estate mock-ups. It went live globally on day one. No waitlist, no vetting, no paid tier.

But why did Google release such a devious feature in the first place? Let's hear it from some experts.

Dutch open-source researcher Henk van Ess put refugees near the Mexican border, planted a nuclear plant in Iran, staged a fatal crash on an Amsterdam street and dropped a bomb crater beside a hospital in Gaza.

Read: How to plant a nuclear plant in Iran | Digital Digging

A fake built inside Google Earth doesn't have to be convincing on its own, either. As van Ess tells BBC Verify, "it inherits the credibility of the map it was born on."

Everyone else who used it found the same thing. NPR generated Iran's Kharg Island oil terminal on fire and a flooded US Capitol.

BBC Verify produced a collapsed Eiffel Tower, a sinkhole swallowing the Great Pyramid of Giza, and Russian tanks in Kyiv. Google insisted its model blocked "harmful topics"; BBC Verify found that slightly vaguer prompts walked straight past the block.

Read: Google withdraws new Earth AI tool after warnings over misinformation risks | BBC

Evan Hill, a visual forensics investigator at the Washington Post, asked how, or whether, this was red-teamed internally, given that "the opportunities for abuse and disinfo are literally boundless."

Investigative journalist Ben Heubl called the launch "absolutely irresponsible and plain bad for investigations." His example is the one that should worry you: type in the address of your workplace, prompt it to look like a drone attack happened there, and you get something that reads like Ukraine, Gaza or Sudan.

The Map Was The Evidence

Google Earth and Street View are not just where you look up a holiday. They are the backbone of open-source verification. Geolocating a video—matching a rooftop, a kerb, a line of trees, the angle of a shadow—frequently ends with a screenshot from Google Earth, placed beside the frame in question, published as the proof.

Those screenshots run through open source investigations, human rights reports, court submissions and have been widely used by pretty much every fact-checking newsroom around the world. In 2014, when the Russian Ministry of Defence produced satellite images about the downing of MH17, Bellingcat checked them against Google Earth's dated archive and showed they were fabricated.

The headline was Who to Trust, Google or the Russian MoD? Back in 2015 that was rhetorical.

In an interview to NPR, Jake Godin, a senior researcher at Bellingcat, explained that he has long treated satellite imagery as one of the few visual references still genuinely hard to fake. Which is exactly why a tool that mass-produces fakes on top of it worries him.

Earlier this year, someone took a genuine Google Earth capture of the US Navy's Fifth Fleet headquarters in Manama, Bahrain, dated 10 February 2025, and pasted AI-generated damage onto it. It circulated as a strike on a facility in Qatar and cleared 950,000 views before it got fact-checked. The forger got sloppy and left the parked cars exactly where they were, right down to the individual vehicles.

That fake took several manual steps to create. Google basically collapsed it into one prompt.

Google says every generated image carries SynthID, checkable through Gemini or Lens. But when Van Ess ran them through Google’s VerifyAI tool, it was unable to find any Synth ID. Hive, one of the biggest AI-detection firms outside Google, also scanned one such visual and rated as, AI Generated Video: 1%,. Deepfake: 0%. BBC Verify separately managed to trick Gemini into calling its own fakes real.

Can Google Put This Genie Back In The Bottle?

Google pulled the feature in under 48 hours. Its own rollback statement conceded that people trust Google Earth as a reliable view of the world, and that screenshots of generated images had been circulating in ways that appeared to breach its policies. The pause is framed as temporary, "while we work on implementing stronger guardrails."

Read Google's rollback statement

What that actually says is not "we should not have done this", but rather "we will do this again, better."

Google's other defence? That generated images never appeared inside the shared map for other users. However, screenshots could be easily taken, and this would also suggest that the credibility of Google Earth screenshots would be severely compromised.

Pulling the feature doesn't undo the harm it has done already.

We Cannot Drink DATA

Last month I wrote about how data centres blocked by American communities are quietly reappearing in India, and about the deal that made it worth everyone's while: a tax holiday until 2047 for foreign cloud firms on income earned from customers abroad, so long as it's routed through data centres on Indian soil.

In Andhra Pradesh, the state threw in 480 acres and an incentive package worth Rs 22,002 crore over 20 years, with no cumulative environmental impact assessment made public and a regulatory classification that conveniently skipped mandatory public hearings.

Reuters visited the Google–Adani site outside Visakhapatnam this week and found a “hillside stripped to red earth and terraced into steps”, earthmovers still working. They also found activists and schoolchildren marching through the city with banners reading "We cannot drink DATA," painting handcuffs onto the Google logo.

Read: Google's $15 billion India data centre project battles water, wildlife concerns | Reuters

Visakhapatnam takes in 410 million litres of water a day against a requirement of 480 million. Rationing is routine for its 2.5 million residents. Regardless, the state has promised the project a "guaranteed" supply for 20 years anyway.

The Kambalakonda Wildlife Sanctuary, home to leopards and pangolins, sits 860 metres from the construction site. The state says that's farther than the law requires, while Google says it will rely on advanced air cooling to protect local water resources and install sound-dampening measures at the site.

Four legal challenges are now live: a public interest litigation before the Andhra Pradesh High Court filed by Jal Biradari, next heard on 24 August, and three more at India's environmental court brought by the Human Rights Forum, arguing the state cleared the project without assessing its impact on a rural drinking-water scheme.

In July, Americans held 142 protests across 42 states against data centres, and some of them won. In Vishakhapatnam, the hillside was already terraced before the first hearing date was even set.

MESSAGE FROM OUR SPONSOR

Own Search With Podcasts

Your competitors are fighting over the same keywords. The smartest brands are building the authority that search engines, AI platforms, and customers trust everywhere.

Every relevant podcast appearance can produce branded mentions, backlinks, transcripts, citations, clips, expert content, and third-party proof that keeps compounding across search and AI discovery.

PodPitch searches millions of podcasts, finds the shows that matter to your market, develops the angle, sends personalized pitches, and follows up automatically until your experts are booked.

Growth teams are already using podcast appearances to build distributed authority that cannot be manufactured by publishing another generic SEO article.

Only 20 SEO, AEO, and GEO demo spots are available this month. Once they’re claimed, the offer disappears.

Start building searchable authority now, before your competitors own the conversations shaping your market.

Appear on 3,853,234 Podcasts

📬 READER FEEDBACK

💬 What are your thoughts on using AI chatbots for therapy? If you have any such experience we would love to hear from you.

Share your thoughts 👉 [email protected]

Have you been targeted using AI?

Have you been scammed by AI-generated videos or audio clips? Did you spot AI-generated nudes of yourself on the internet?

Decode is trying to document cases of abuse of AI, and would like to hear from you. If you are willing to share your experience, do reach out to us at [email protected]. Your privacy is important to us, and we shall preserve your anonymity.

About Decode and Deepfake Watch

Deepfake Watch is an initiative by Decode, dedicated to keeping you abreast of the latest developments in AI and its potential for misuse. Our goal is to foster an informed community capable of challenging digital deceptions and advocating for a transparent digital environment.

We invite you to join the conversation, share your experiences, and contribute to the collective effort to maintain the integrity of our digital landscape. Together, we can build a future where technology amplifies truth, not obscures it.

For inquiries, feedback, or contributions, reach out to us at [email protected].