AI glossary · Beginner · AI security · Generative media
Deepfake
A deepfake is an image, video or audio clip in which AI realistically synthesizes or swaps a real person's face, voice or actions. The word combines deep learning and fake.
About this term
- Level
- Beginner · Core ideas for first-time readers
- Category
- Basics
- Fields
- AI security, Generative media
- Added
- 2026-10-09
How it works
An AI model sees and hears many photos or recordings of a specific person and learns the features of their appearance and way of speaking. It can then replace the face in another video with that person's face, or make any sentence be read in that person's voice. Early on, generative models such as GANs were mainly used; now, advances in diffusion models and TTS (speech synthesis) have made the quality much better. A major change is that convincing results can now be made from less material than before.
Why it matters
Deepfakes can be useful in film production, dubbing, and restoring the voices of people who have lost them. But they also cause serious harm, such as money transfer scams that imitate the voice of a family member or executive, non-consensual sexual imagery, and fake videos of politicians' statements. Above all, they are a problem because they shake social trust in whether 'what we see and hear can be believed'. Another side effect is that even genuine videos can be denied by claiming they are deepfakes.
Things to know
Detection technology for spotting deepfakes is also advancing, but because it is locked in a cat-and-mouse race with generation technology, perfect detection is difficult. So alongside detection, methods that 'mark content as genuine from the start', such as AI watermarks and content provenance, are being researched. For individuals, the most practical defense is the habit of re-verifying through a different channel when receiving a suspicious contact, such as an urgent request to transfer money. Many countries are putting in place regulations that punish creating and distributing non-consensual deepfakes or require labeling of AI-generated content.
Example
As scams became known in which a deepfake imitating a company executive's face and voice in a video conference call instructs an employee to transfer money, many companies are changing their rules so that large transfers are not processed on video or voice instructions alone but are re-verified through a separate procedure. In the video production industry, on the other hand, the same technology is sometimes used, with actors' consent, to dub into other languages with matching lip movements.