Explainer · Safety
That video of the candidate may be fake. Spotting it no longer depends on your eyes
- Written by
- Equipo CIUDADAN-IA
- Translated by
- Equipo CIUDADAN-IA
- Validated by
- Equipo CIUDADAN-IA
The short answer
With AI, videos and audio are now fabricated in which a real person says things they never said. They are called deepfakes and they have already circulated in Latin American elections. The bad news: the human eye is no longer enough to detect them. The good news: habits do work — distrust what arrives through forwarded chains, look for the original source, and wait before sharing. Lies travel faster than corrections; your pause is the brake.
In depth · 2 min read
What exactly is a deepfake
A deepfake is a video, audio clip or image fabricated with AI that shows a real person doing or saying something that never happened. The technology learns from authentic recordings — a few minutes of voice or a few dozen photos are enough — and then generates new content with that face and that voice. What required a Hollywood studio five years ago is now done with an app.
Why it matters in an election
Election season is perfect terrain: emotions run high, there is little time to verify, and many people are sharing. A fake audio of a candidate “admitting” fraud, released the night before the vote, can travel through WhatsApp to millions of phones before any rebuttal exists. Several countries in the region have already documented such cases — cloned audio, manipulated video, fake news bulletins read by synthetic voices.
There is a second, quieter harm: once people know deepfakes exist, any politician can dismiss a real, inconvenient video by saying “that’s AI.” The lie wins twice: when the fake is believed, and when the truth is doubted.
What no longer works
For a while, tricks were taught: watch the blinking, the hands, the edges of the face. They are no longer reliable — the models improved. Automatic detectors are no guarantee either: they succeed sometimes, fail other times, and always arrive late.
What does work
The defense is habits, not technology. First: look at the channel. Was it published by an identifiable outlet, or did it arrive forwarded with no origin? Content that exists only in messaging chains deserves maximum suspicion. Second: find the original source — if the candidate really said it, a serious outlet will have it. Third: wait. Manipulation bets on your immediate reaction; sharing an hour later, once verified, takes no force away from the truth, but it does from the lie. And fourth: when in doubt, don’t share — not spreading something false is also participation.