A deepfake is synthetic content—video, image, or audio—generated or manipulated using artificial intelligence techniques, typically generative adversarial networks (GANs) or diffusion models, to hyper-realistically mimic a person's face, voice, or gestures. The result can show someone saying or doing things that never happened, with a degree of plausibility that makes it difficult to distinguish from authentic material.
Its importance is twofold. On one hand, it offers legitimate uses in cinema, dubbing, accessibility, or audiovisual restoration. On the other hand, it poses serious risks:
- Disinformation and political or media manipulation.
- Identity theft in fraud and scams (for example, voice cloning to authorize payments).
- Privacy violations, especially with non-consensual sexual content.
In practice, detecting them is increasingly complex, although tools exist that analyze inconsistencies in blinking, lighting, or digital artifacts. The regulatory trend, such as the EU AI Act, requires labeling this type of content to protect the user.