Definition
Deepfake is a deepfake is a video, image or audio clip made or altered by AI so that a real person appears to say or do something they never did, often convincingly enough to fool people who know them.
A deepfake is synthetic media built to look or sound like a specific real person. The name blends "deep learning," the technique behind it, with "fake." The result can be a face swapped onto another body, a lip-synced video of someone reading words they never said, or an audio clip in a cloned voice.
Most deepfakes start with examples of the target: photos, video or recordings. A model learns the person's face, voice and mannerisms from those examples, then generates new content that matches them. Early tools relied on a pair of networks that competed, one producing fakes and one trying to spot them. Newer systems often use a diffusion model, which builds an image or video by gradually removing noise until a convincing picture emerges. Audio fakes usually rely on voice cloning.
Think of it like a skilled impersonator who has studied hours of footage. The difference is that software never gets tired, and it can be pointed at anyone with enough public video.
Deepfakes are used for fraud, harassment, political misinformation and fake endorsements. The fraud case is the one that costs businesses money: a caller who sounds like your boss, or appears on a video call, asks for an urgent payment. People trust faces and voices by instinct, and that instinct is exactly what the technique targets.
They also have legitimate uses, such as film effects, dubbing a video into another language, or giving a voice back to someone who lost theirs. The technology is neutral. The consent of the person being imitated is what separates the uses.
Detection tools exist, but they are in a constant race with the generators, so treat them as a clue and not a verdict. Habits work better than software:
On the technical side, AI watermarking hides a signal in generated content, and C2PA attaches signed records of where a file came from. Both help prove what is real, but only when the creator applies them. A related idea, proof of personhood, tries to confirm that a real human is on the other end.
Not every AI-generated image is a deepfake. The word applies when the output imitates a real, identifiable person without that person being shown as the source. A fantasy landscape from generative AI is not one. A fake video of a named executive is.
Look for odd blinking, blurred edges around the face, mismatched lighting or lip movement that lags the audio. None is proof, so verify important requests through a second channel.
It depends on the use and place. Fraud, harassment and some election or intimate-image uses are illegal in many jurisdictions, while parody and licensed uses can be legal.
World launched World ID for banking and finance at Money20/20 on Sunday, bundling human verification, Zoom checks and AgentKit, a beta that lets an AI agent prove it acts for a real person. No bank has signed on publicly yet.
Federal prosecutors charged three people with using AI to mimic CEOs' voices and faces on calls, taking at least $129 million from companies between July 2024 and February 2026. If your finance team treats a video call as proof, that habit is now the hole.