Deepfakes are AI-generated or AI-altered images, video, and audio realistic enough to pass for real. What once needed a studio now needs a laptop and a few minutes.
How they're made
Powered by diffusion models and GANs for faces, plus voice-cloning models that mimic anyone from seconds of audio. Face-swaps, lip-sync to fake speech, and full synthetic people are all routine now.
The harms
| Use | Harm |
|---|---|
| Non-consensual imagery | The most common abuse, overwhelmingly targeting women |
| Fraud | Cloned-voice "family emergency" and CEO scams |
| Political | Fake clips to sway elections (AI and Misinformation) |
| "Liar's dividend" | Real evidence dismissed as "probably a deepfake" |
The deepest damage isn't any single fake — it's the erosion of a shared baseline of truth. When anything could be fake, nothing has to be believed.
Detection & defenses
- Detection models (an arms race — generators keep improving).
- Provenance standards like C2PA that cryptographically sign authentic media at capture.
- Media literacy and verification habits.
Related: AI and Misinformation · Diffusion Models Explained · Facial Recognition and Surveillance