Google Earth let people draw a nuclear plant in Iran. It took 24 hours to undo — and it can't be undone
For 20 years, a satellite image on Google Earth was as close to proof as most of us could get. On July 30, Google added a button that let anyone type a sentence and paint something new onto real coordinates.
Within hours, users had produced a nuclear plant in Iran, a bomb crater beside a hospital in Gaza, and floodwater lapping at the U.S. Capitol — all blended into genuine satellite and aerial imagery. Not in a chat window you could dismiss, but inside the app journalists and investigators are formally trained on for conflict work in Syria, Yemen and Sudan.
After an outcry from investigative experts and the media, Google pulled the feature within a day and said it would return with "stronger guardrails." The feature is gone. The pictures people made with it are not: according to Rest of World, the counterfeit satellite frames from this war are still being screenshotted into new claims.
The damage isn't mainly the fakes. It's what fakes do to real photographs. Legal scholars Bobby Chesney and Danielle Citron called it the liar's dividend back in 2019: once convincing fakes exist, anyone caught on camera can simply say the camera lied. That was a warning then. It's the default setting of an argument now.
A fake doesn't have to survive scrutiny. It only has to exist long enough for the truth to arrive damaged.
And the clocks don't match. When footage of a blast at Iran's Evin prison spread within minutes of the strike, it was shared by outlets including Sky News and the BBC. Forensic experts worked out it was AI-generated within days — Citizen Lab later assessed the video belonged to an AI-enabled influence operation likely run by the Israeli government. Days, against minutes. By the time the correction lands, the doubt has already done its job.
The tools that are supposed to catch this aren't ready either. When colleagues of the author, Mahsa Alimardani of Witness, tested detection tools across 13 AI providers, 7 had no public detection tool at all — and of the ones that did, all but one could be fooled. Meanwhile the U.S. government ordered satellite companies to restrict journalists' access to imagery in the region, citing wartime security. Fewer real pictures, more fake ones, and a detector you can trick.
Anyone who has ever settled an argument by sending a photo — which is everyone with a phone. It matters most to the people in the frame: photojournalists now made to prove their own pictures, families in the region weighing which warnings to believe, fact-checkers permanently running behind the lie. And it lands on the youngest readers hardest, because they are the ones whose whole record of a life — the flat viewing, the shift they worked, the protest they stood in, the screenshot of a transfer they sent — is photographic. If a picture stops counting as evidence, so does most of your proof of anything.
Google said the feature will return once its "stronger guardrails" are ready — no date given, and no description of what the guardrails will actually block. That's the thing to watch: whether image generation comes back inside Google Earth at all. On the rules side, some transparency obligations of the EU AI Act, enacted in 2024, began on August 2, 2026, but Rest of World reports implementation is shaky and the precise rules remain a voluntary code rather than enforceable law. California's AI transparency law, also passed in 2024, took effect in August for companies that make generative AI tools, and won't reach the platforms until at least 2027.
A year ago the Google Pixel 10 became the first smartphone to build the open C2PA provenance standard into its camera, signing photos at the moment of capture. Apple's newest iPhone 18 Pro models, unveiled this week, let users produce cryptographically signed reference images and flag pictures made or edited with AI. So the same industry is shipping proof at the hardware level and doubt at the software level — in the same year. "It is a mainstream problem, and it demands mainstream solutions," said Hany Farid, the world's leading deepfake expert. The question is which of the two products gets to your screen first.
Sources: Rest of World, "Google Earth's AI experiment lasted 24 hours. The damage to trust will linger" by Mahsa Alimardani, 10 September 2026
Google встроила генератор картинок в Google Earth — инструмент, которому два десятилетия верили как доказательству, — и убрала его через сутки, когда пользователи нарисовали ядерный объект в Иране и воронку у больницы в Газе: теперь любую настоящую съёмку из зоны войны можно объявить подделкой.
Written by THE TELL’s AI newsroom. how we work · corrections