Google pulls its Earth AI image tool after it produces fake satellite images in a day

[Inference Reuben]

Google pulled its new AI image generation tool from Google Earth just one day after launching it, and honestly the reason why is a pretty stark illustration of how badly this could have gone

The feature let anyone zoom into any location on Google Earth, hit create image, and type a prompt to generate a photorealistic image layered directly onto the real satellite and 3D mapping data for that spot, powered by Googles Nano Banana 2 model, and the spatial anchoring is what made it so convincing since a fake flood or fire would appear at the exact real coordinates with accurate building positions and scale

Researchers and journalists wasted no time testing the limits, generating fake flooding in Washington DC, fires at an Iranian oil terminal, a fabricated nuclear site, a bomb crater in Russia, and even a homeless encampment on the White House lawn, all of which looked disturbingly plausible at first glance

Googles main defense was that every generated image carried a SynthID digital watermark that could be checked through the Gemini app or Lens in Search, but critics quickly pointed out two problems, most people never bother verifying an image before sharing it, and Googles own Gemini chatbot reportedly failed to reliably identify some of the AI generated fakes when tested directly

Open source intelligence analysts who rely on Google Earth as a trusted record for monitoring conflict zones and disasters were apparently especially alarmed, since the whole value of that kind of imagery depends on it being an authentic record rather than something anyone can fabricate on command

Google says the images never appeared in the main Google Earth experience for other users and only spread through screenshots shared elsewhere, but that framing feels like a pretty thin defense given how fast fabricated satellite images already circulate online during real conflicts, this pull happening within 24 hours of a global no waitlist launch says a lot about how badly this was underestimated internally

Google withdraws Earth AI tool after misinformation warningswww.bbc.co.uk

[Christopher]

The fact that their own watermark detection system couldnt reliably catch the fakes is the real story here, that was supposed to be the entire safety net and it just didnt hold up

[ForumGremlin93]

Google Earth is one of the last tools OSINT analysts trust precisely because its supposed to reflect reality, adding a fake image generator into that same product was a genuinely baffling decision from the start

[ThreadNecro11]

Anchoring the fake image to real coordinates and real building footprints is what makes this so much scarier than a normal AI image generator, it inherits the credibility of the real satellite data underneath it

[Freddy93]

This is basically a preview of what state actors are already doing manually with fake satellite images during conflicts, Google just handed everyone an easier tool to do the same thing instantly

[MiniElliot]

Real question is whether this comes back later with better safeguards or if the reputational damage was enough that Google just quietly shelves the whole idea long term

[Matthew51]

The SynthID watermark defense always felt weak to me even before this, most people sharing a shocking image online arent stopping to open Gemini and check its authenticity first

[IronMan70]

Kind of wild that they launched with no waitlist and no restrictions at all given how obviously this could be abused, feels like a rushed decision from product leadership over actual safety review