deepfake
A deepfake is an image, audio, or video created or altered with artificial intelligence that appears deceptively real, typically showing a real person in a situation that never occurred.
In plain terms
Imagine the world's most skilled voice imitator and mask-maker in one person, who can impersonate anyone from a few photos and a few seconds of audio. A deepfake does the same, but with software: it learns a person's face, expressions, and voice, then creates a recording in which that person says or does something they never actually did.
The technique itself is not bad. In films, it is used to de-age actors, in dubbing to match lip movements to translated text, and it can also restore the voices of patients who have lost theirs.
The problem begins when it is used for deception: fraud, political smear campaigns, or intimate recordings made without someone's consent. What distinguishes it from old photo montages is that today no expertise is needed, it can be created in minutes, and it works with moving images and live calls.
How it works
- Collection: the creator gathers images, videos, or audio recordings of the target person, often simply from social media. A few seconds may suffice for voice cloning today.
- Learning: an AI model learns the shape of a face, facial expressions, voice timbre, and speech rhythm from patterns. Modern tools often build on pre-trained models, so few examples are sufficient for them.
- Generation: the model produces new content. It places a face onto another person's body, adjusts lip movement to match given text, or has something read aloud in the cloned voice.
- Refinement: the software smooths out telltale errors, such as lighting, facial edges, and blinking, to make the final result look natural.
- Distribution: the finished material goes to social media, a messaging app, or a live video call, where people mostly react quickly without verification.
An everyday example
Your 'grandchild' calls with a crying voice, saying they were in an accident and you must transfer money immediately. The voice sounds exactly like theirs because it was cloned from a public video, but a fraudster is on the other end of the line. In such cases, hang up and call them back on their usual number.
Who is affected
A common misconception
Limits and risks
- Detection tools are not infallible: they often lag behind the tools that generate fakes and may flag real recordings as fake.
- If someone asks for money, a password, or an urgent decision during a call or video, always verify via another reliable channel, such as a callback or a pre-agreed family password.
- The more public images and audio recordings available about you or your child, the easier it is to create a forgery from them.
- Because of deepfakes, genuine recordings can also be dismissed as fake. A video therefore proves less and less on its own if we do not know where it comes from.
Similar, but not the same
- fake news – Fake news can be any form of false claim, while a deepfake is specifically an AI-forged image, audio, or video often used to support fake news.
- photo montage – The montage is assembled by hand by a person in an image editor, while the deepfake is generated largely automatically by AI and works in video and audio as well.
- AI-generated image – An AI image of a fictional landscape or a non-existent person is not a deepfake, because that term refers to imitating a real person or event with deceptive fidelity.
Related concepts
voice cloning, arccsere, lip-sync forgery
watermarking, phishing, MI-rendelet, deep learning, content authentication
A brief history
The word is a portmanteau of "deep learning" and "fake". It spread in late 2017 when a Reddit user named "deepfakes" posted videos of celebrities with swapped faces. These early face-swaps were created mainly using so-called autoencoder networks. Another key milestone in the field was the generative adversarial network (GAN) introduced in 2014, which advanced realistic image generation.
Where things stand
By the end of 2026, anyone can create convincing fake audio and video in minutes with cheap tools, and scammers do so regularly. The EU AI Act requires anyone who publishes a deepfake to disclose that the content was artificially generated or manipulated. However, detection and enforcement of the rules in practice lag behind the pace of the technology.
In the news
- 88-year-old loses $450,000 in deepfake scam; bank will not reimburse under Canadian rulesNov 8, 2026 · Money
- AI sex ads and deepfake micro-dramas appear in children's feeds; TikTok says they are legalNov 4, 2026 · Marketing
- AI-generated fake images target students and teachers as schools remain unpreparedNov 4, 2026 · Education & Family
- Hungarian professional publications: shadow AI, deepfakes and data protection risks at companiesNov 3, 2026 · Jog
- Delhi High Court orders removal of deepfake videos and protects personality rightsNov 2, 2026 · Jog
- Criminal-law responses to deepfake abuse: Australia's first conviction, Queensland proposal, Irish billNov 2, 2026 · Jog
- Only one in five people in Hong Kong trust banks to stop deepfake scamsNov 1, 2026 · Money
- Cybernews and UNESCO: AI deepfake abuse targets girls in schoolsOct 27, 2026 · Education & Family
Also known as
deepfake · deep fake · deepfake · deepfake · deepfake video · deepfake hang
Sources
This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.