Deepfake

A deepfake is synthetic media convincing enough to pass as real: a politician’s cloned voice, a celebrity’s face swapped into a video, a fake photo of an event that never happened. The term comes from “deep learning” + “fake” and appeared around 2017.

How they’re made

Generative SI, especially diffusion models and voice-cloning systems, can now produce realistic faces, voices and video from a few seconds of source material or a text prompt.

Harms

• Fraud: cloned voices of relatives or executives used to demand money
• Non-consensual intimate imagery, overwhelmingly targeting women and girls
• Political disinformation, especially around elections
• The “liar’s dividend”: real evidence dismissed as fake

Responses

• Laws: the U.S. TAKE IT DOWN Act (2025) criminalises publishing non-consensual intimate deepfakes and requires platforms to remove them; many states have election and impersonation laws. The EU AI Act requires deepfakes to be labelled.
• Technology: watermarking and content-provenance standards such as C2PA, plus detection tools, though detection remains unreliable.
• Habits: verify surprising media through a second source, and agree on a family “safe word” for urgent phone calls.