The dangers of deepfakes

What are deepfakes?

Deepfakes are synthetic media in which AI realistically imitates the face, voice, or body of a person in a video, photo, or audio recording. The name is a combination of 'deep learning' and 'fake'.

The technology has improved dramatically in recent years. What previously required hours of computing power and technical expertise can now be done with a smartphone app in minutes.

Illustration created with Canva AI

How do deepfakes work?

Deepfakes use Generative Adversarial Networks (GANs) or diffusion models. A generator creates fake synthesis; a discriminator tries to distinguish fake from real. By repeating this game, the fake images become increasingly realistic.

For a convincing deepfake, you need original footage of the target person — the more the better. Public figures with many online videos are therefore most vulnerable.

The dangers

• Disinformation — fake videos of politicians saying things they never said
• Election manipulation — a fake speech by a candidate, spread just before the vote
• Fraud — CEO fraud via a fake voice instructing employees to transfer money
• Intimidation — non-consensual intimate imagery (NCII): pornographic deepfakes of private individuals
• Reputational damage — placing someone in compromising situations that never occurred

Detection

AI detection tools look for inconsistencies: strange eye movements, odd hairlines, inconsistent lighting. But as deepfakes improve, detection relatively worsens. It is an arms race.

What can you do?

• Be skeptical of shocking video evidence — check the source
• Companies: train employees to recognize deepfake fraud
• Governments: the EU AI Act prohibits certain applications of deepfakes; more legislation is coming