Deepfake

A deepfake is a "fake that looks real," made with AI. The word combines "deep learning" and "fake," and it covers everything from face-swapped videos to cloned voices to speeches showing someone saying things they never said.

The technology itself is neutral. The same tools are used in movies to de-age actors or to recreate the stage performance of a deceased singer. The problem arises when it's used without consent, to deceive — celebrity-impersonation investment scams, sexual content made by grafting an acquaintance's face onto it, fake speech videos released right before elections. These are recurring incident types, not hypotheticals.

That's why most deepfake news comes paired with a "response" angle: invisible watermarking technology embedded in AI-generated content, platform labeling requirements, and punitive legislation in various countries. It's more accurate to read these as social and legal stories than as tech stories.

"Deepfake videos of a candidate spread ahead of the election — platforms on high alert." This is the term's typical use, where the article centers on verifying authenticity and regulatory response.

1. When an image on social media looks suspicious, zoom in on: the number and joints of fingers, text in the background (signs, banners), symmetrical items like earrings or glasses, and the boundary between hair and background.
2. For video: check whether blinking looks unnatural, whether the edges of teeth look smeared, and whether the voice is subtly out of sync with the mouth movements.
3. The most powerful tool is reverse image search — search the image with Google Lens to trace its original source. Detection tricks get defeated as the technology improves, but "checking the source" never does.