Glossary
Deepfake
A deepfake is AI-generated audio, video or imagery that convincingly imitates a real person: their face, their voice, their gestures. Some uses are harmless entertainment; the dangerous ones power scams and disinformation, which is why the EU now requires labeling.
AI can now clone a voice from a short audio sample and put words in the mouth of someone on video. That’s a deepfake. The technology itself is neutral, and film studios use it legitimately, but it has given old scams a new engine.
The scenario worth knowing, told calmly: a grandmother gets a phone call. It sounds exactly like her grandson, his voice, his way of speaking, saying he’s in trouble and needs money fast. The voice is cloned from clips found online. The protection isn’t technical, it’s procedural: hang up and call the real person on their usual number, or agree on a family code word in advance. Urgency plus a request for money or codes is the signature of the scam, however convincing the voice.
The law is catching up. The EU AI Act requires AI-generated and manipulated content to be disclosed, and providers are rolling out AI watermarking to make machine-made media detectable. Neither is foolproof yet, so healthy verification habits remain your best layer.
Where you’ll meet this
- “AI-generated” labels on images and video on Meta, YouTube and TikTok
- Voice-cloning warnings from consumer protection agencies and banks
- Detection and watermarking notes published by AI vendors