Deepfake detection is software that decides whether an image, a voice recording or a video was made or changed by AI, or whether it was ever real. A deepfake is media in which AI has created a face, a voice or a whole person, or put one person's face or voice onto another. Such media is also called synthetic media. A platform, a newsroom or a bank's verification team uses deepfake detection before it trusts what it sees or hears.
How does it work? A detection model is a kind of AI model trained on a large number of real and fake examples. It learns the small traces that generating software leaves behind. In a video, these traces may be in the edges of the face, the lighting, the blinking or the way the lips match the sound. In audio, they may be in the breathing, or in frequencies that a human throat does not produce. The model does not give a yes or no. It gives a score, such as "87% likely to be synthetic", and marks the parts of the file that look faked.
Imagine the fact-check desk of a news website in Kolkata, a week before an election. A 40-second audio clip is spreading on chat apps. In it, a candidate seems to announce that he is withdrawing from the race. The desk uploads the clip to a detection tool. The tool scores it 92% likely to be synthetic, and marks two sections where the voice changes in a way that real speech does not. The desk does not rely on the score alone. It also asks the candidate's office for a statement and looks for the original recording. Then it publishes its finding.
Banks and fintechs use the same tools during video KYC, the account-opening video call. A fraudster may play a deepfake video into the camera instead of showing a real face. Liveness detection checks that a living person is in front of the camera. Deepfake detection checks whether the video itself was generated. Both feed into identity verification (IDV).
Social apps use it on reports. When a trust and safety investigation receives a suspect photo or voice note, it can send the file for a score.
Not all synthetic media is harmful. Studios use AI dubbing and lip sync so that a film's actors appear to be talking in another language. So a high score does not by itself mean that someone broke a rule. A person must still judge what the media is used for.
Detection is a contest. Each new generator makes fakes harder to spot, and each new detector learns from them. This is described in AI in cybersecurity.