Google Earth’s AI Editor Lasted a Day Before Deepfake Fears Won

Google Earth’s AI Editor Lasted a Day Before Deepfake Fears Won
Google’s pitch was imaginative: use AI to visualize a rebuilt historic site or a planned community court. Its critics saw something far more consequential — a machine for putting convincing lies on top of a map people trust.

On July 30, Google promoted Nano Banana 2’s arrival in the web version of Google Earth, saying users could combine the model with the platform’s satellite, aerial and 3D imagery. A post amplified by Google DeepMind chief Demis Hassabis framed it as a way to see a city a century ago or picture a new local basketball court.

The appeal was that the generated scenes were grounded in real locations. That was also the problem. Google Earth is routinely used by journalists, researchers and open-source investigators to check where images were taken. Researchers swiftly tested the new prompt-based editor for the inverse purpose: manufacturing visual evidence.

Henk van Ess said he typed a single sentence and placed “refugees near the Mexican border,” then added a nuclear plant in Iran and a fatal crash in Amsterdam, all over Google’s own imagery. Other demonstrations included a bomb crater near a Gaza hospital and a huge golden Donald Trump statue over the White House. The concern was not merely that images could be fabricated, but that a trusted verification tool could become an alibi for fabrications — or for dismissing authentic satellite evidence as AI.

Google initially pointed to SynthID watermarks, Gemini and Lens checks, saying it took misinformation seriously and blocked harmful requests. But van Ess argued that fakes rarely circulate with intact metadata: they move as screenshots, re-encodes and phone-recorded clips. Testing also suggested generated material could evade at least one AI detector.

Within a day of the rollout, Google withdrew the feature. It acknowledged useful professional applications but said screenshots of generated imagery appeared to violate policy, adding: “We’re rolling back this feature in Google Earth while we work on implementing stronger guardrails.” Google stressed that the images were watermarked and did not appear in the main shared Earth experience.

One report described a separate pause after synthetic imagery, including a fake image of Pope Francis, underscoring the same wider problem: once realistic AI images spread, provenance becomes harder to recover. Google’s retreat resolves the immediate flaw, not the larger contest over whether seeing is still believing.

Continue reading https://foxvector.com/stories/019fd199-0c9d-0dbb-7265-3b9af270a4ee