A deepfake is a fake video, photo or audio made with AI to look real: someone's face on another person's body, or their voice saying something they never said. The name blends deep learning and fake. It shows up in comedy and film, and also in scams.
A question from the app. Pick an answer.
Your brother calls, in his voice: he is in trouble and needs money now. What do you do?
A photo or video in the same chat is not enough to verify who it is: AI can generate those too.
A voice can be cloned from a short clip. Calling the number you already had does not depend on spotting the fake.
A cloned voice can sound just the same, and flaws are harder to hear on the phone. How it sounds no longer proves who it is.
A deepfake comes from generative AI: a neural network learns from photos, videos or audio of a person and generates new content with their face or voice. The most common cases are a face swapped into a video, a cloned voice and an invented photo of something that never happened.
A short clip is enough to clone a voice, like one from a posted video or a voice message. And over the phone the fake is harder to spot.
In video: a mouth that does not quite match the words, odd teeth or hands, blurry edges around the face, light that does not match the background, skin that is too smooth. In audio: an even voice with no breathing, pauses in odd places or an emotion that never shifts.
They help, but less and less: models keep improving and a good deepfake may have none of them. Not spotting anything odd does not prove it is real.
The classic case: your daughter calls, in her voice, crying. Something happened, she needs money now and asks you not to tell anyone. Urgency, secrecy and money together are the alarm, however real it sounds.
You do not need to beat the deepfake by looking: hang up and call the number you already had, or message them on another app. If they do not answer, ask someone who is with them. A photo or video in the same chat is not enough to verify who it is: those can be generated too.
One more trick: agree on a family code word that is not written anywhere. If the caller does not know it, do not send money: verify through another channel. The same goes at work for an odd request from your boss on a video call: confirm it through the usual channel.
"If I see it on video, it is real." Not anymore: a video or a video call can be faked too.
"A deepfake detector will settle it." They help, but they get it wrong both ways. Treat the result as one more clue, not a verdict.
"Everything fake is a deepfake." A real photo taken out of context or a trimmed clip fools people just as well, no AI needed. The question is the same: where did it come from and who confirms it?