What Is a Deepfake? Synthetic Media and AI Impersonation Explained

What is a deepfake?

A deepfake is synthetic or manipulated media, audio, video or an image, that has been generated or altered by AI to appear authentic. The umbrella term is synthetic media. A deepfake is the deceptive use of it: media engineered to pass as real so it can impersonate a person, move money, damage a reputation or seed a false narrative.

The word that matters is deceptive. A visual effect in a film is synthetic media, and no one is fooled. A cloned voice of a chief executive authorising a payment is a deepfake, because it is built to deceive.

How deepfakes are made

Modern deepfakes come from generative models that learn the statistical patterns of real media and then produce new media that shares those patterns. Diffusion models and GANs generate imagery; voice-cloning models reproduce a specific person’s speech from a short sample. The common forms are:

• Face swaps, where one person’s face is mapped onto another’s body in video.
• Full-face synthesis, where an entirely new, non-existent face is generated. A fresh face from a service such as thispersondoesnotexist.com is a simple example.
• Lip-sync reanimation, where a real clip is re-driven so a person appears to say words they never said.
• Voice clones, where a few seconds of audio is enough to synthesise new speech in someone’s voice.

None of these require a specialist. The tools are cheap, fast and widely available, which is why the volume of synthetic media has risen sharply.

Why they are hard to spot

The perception gap has closed. Modern synthetic media is, in many cases, indistinguishable to the human eye and ear. Looking for obvious visual flaws is no longer reliable, because the flaws have largely gone.

Provenance signals help only when the generator cooperates. Content credentials such as C2PA and watermarking approaches such as SynthID can mark media as AI-generated at the point of creation, but most malicious media carries no provenance at all: it is stripped, re-encoded or produced by a model that never added it. So detection cannot depend on metadata. It has to read the media itself.

Where deepfakes show up

For organisations, deepfakes cluster around a few high-stakes uses:

• Coordinated narrative attacks and disinformation, where synthetic media is manufactured to make a false story look credible.
• Brand and executive impersonation, where a synthetic clip or cloned voice of a leader is used to mislead staff, customers or markets.
• Financial fraud, where fabricated media pressures a person or a process into moving funds.

The common thread is not the media on its own. It is the intent to deceive at scale.

The coordination connection

A deepfake rarely travels alone. On its own, a single synthetic clip is inert; it does damage when a network of accounts amplifies it fast enough to manufacture the appearance of organic momentum. Detecting the synthetic media answers one question, is this real. Detecting the coordinated inauthentic behaviour pushing it answers the other, who is behind this, and are they acting together.

That is why content authenticity and behavioural detection belong on the same platform. A deepfake used in a narrative attack is both a fake artefact and a coordinated campaign, and you need to see both halves to respond.

How detection works, and where to go next

Detection models do not hunt for visible mistakes. They look for the statistical traces that generative models leave behind, and they combine several models across audio, image and video for reliability. To understand the science, read how deepfake detection works.

To see authenticity applied to a real problem, protecting named leaders from synthetic impersonation, see executive protection. And to see authenticity sit alongside behavioural and coordination signals in one platform, see Signal by AI Uniti.