Deepfake

A deepfake is audio, an image, or video created or manipulated using artificial intelligence to present realistic-looking or realistic-sounding content that can be mistaken for an authentic recording.

More than a face-swapped video

Deepfakes can alter a face, fabricate speech, or create other convincing media. Voice cloning reproduces a particular person’s voice; deepfake covers both audio and visual media.

An ordinary edit or a real recording taken out of context can also deceive people. “Deepfake” specifically refers to AI-created or AI-manipulated media, not every misleading image or recording.

NSA, FBI, and CISA guidance describes threats such as impersonating leaders and financial officers, fraudulent communications, and damage to brands. For a security team, the practical question is what someone wants the media to prove: who they are, what instruction was given, or whether an event happened.

Detection does not replace approval

GAO’s 2024 review describes limits to deepfake detection in real-world conditions. Recording quality and differences from training data can affect a detector’s result. Consider that result alongside the source and context.

For a sensitive request, confirm the requester independently and follow the approval process. Even a genuine video of an executive does not authorize every payment or data release.

See how to verify a request made in a familiar voice for a practical example.

Sources

See it in practice