Deepfakes and Synthetic Media

A video circulates online. In it, a well-known politician appears to announce a shocking policy reversal — one they never actually made. The video looks completely real: the lighting is natural, the voice matches, the facial movements are convincing. Millions of people watch it before fact-checkers confirm it was entirely artificial. This kind of content is called a deepfake — a video, audio recording, or image in which a real person's likeness has been convincingly faked using AI tools. Deepfakes are one of the most consequential applications of generative AI, and understanding them is now a basic media literacy skill.

How Deepfakes Are Made

Deepfakes are created using a class of AI models called generative models — systems trained to produce new content, such as images, audio, and video, rather than just classify or analyze existing content. The original deepfake technique used a type of system called a generative adversarial network, or GAN. A GAN trains two neural networks against each other: a generator that tries to create convincing fake images, and a discriminator that tries to spot the fakes. Through repeated competition, the generator gets better and better at producing images the discriminator cannot detect. The result is synthetic media that looks real. Modern systems use additional techniques including diffusion models — a different AI architecture — and voice cloning software. Together, these tools allow someone to take a few minutes of real video or audio of a person and generate convincing fake footage of them saying or doing things they never said or did.

Why Deepfakes Are Dangerous

The danger of deepfakes is not just that they exist — it is that they erode the shared sense of what is real. Before deepfakes, video was generally treated as strong evidence. Seeing someone in a video meant something. Deepfakes break that assumption. Once people know that convincing fake video exists, they begin to doubt real video too. This creates a difficult situation: legitimate footage of real events can be dismissed as a deepfake, while fake footage of invented events can be believed as real. Researchers call this the liar's dividend: the benefit bad actors gain when deepfakes are so common that people doubt all video evidence, including genuine recordings. Political figures can deny authentic videos of themselves by claiming deepfake. Wrongdoers can escape accountability by casting doubt on real evidence.

Synthetic Media Beyond Deepfakes

Deepfake video is the most dramatic form of synthetic media, but it is not the only one. The same AI advances have produced a wider ecosystem of synthetic content. AI-generated images: tools like image generators can produce photorealistic images of people, places, and events that never happened. A fake photo of a famous person at a fake event can spread on social media before anyone investigates. Voice cloning: AI can clone a person's voice from a few seconds of audio. Scammers have used voice clones to impersonate family members in emergency calls, convincing victims to wire money. AI-written text: while not visual, AI-written articles, fake reviews, and social media posts are a high-volume form of synthetic content that spreads misinformation at scale. Synthetic identities: entirely fake people — with generated photos, generated biographies, and generated social media histories — are used to build fake influence online.