Black October, explainedAppsGuidesAI DictionarySmall FixesOn RecordAboutFollow the lab

Deepfake

Also called deepfakes, face swap

A deepfake is a video, photo or voice clip made or changed with AI so that a real person seems to do or say something they never did.

Images, voice and videoEveryday

In plain words

You get a forwarded video of a well-known actor promoting an investment scheme. Their face, their voice, their way of talking. But they never said it. The clip was made with AI. That is a deepfake.

The name joins "deep learning", the kind of AI used, with "fake". A model studies many pictures or recordings of a person until it can produce new ones: a face placed onto someone else's body, lips moved to match new words, or a voice reading a script the person never saw.

Deepfakes have harmless uses, such as film effects and dubbing. They are also used for scams, false news and harassment. The best defence is a habit: pause before sharing, check whether a trusted source reports the same thing, and look for signs like odd blinking, blurry face edges or lips that drift from the sound.

Everyday example

A forged signature on a cheque

A forger practises someone's signature from old cheques until it looks right at a glance. A careful bank clerk does not just glance: they compare it with the record, check the stroke and call the account holder if something feels off. Deepfakes are forged faces and voices. Your defence is the clerk's habit: look closely, and verify through another channel.

Where the example breaks: A signature check can be done by eye. The best deepfakes now fool trained people and even detection tools, so the visual signs are hints, not proof. Checking the source matters more than spotting a glitch.

Go deeper

How it actually works

Deepfakes are a use of Generative AI. The methods have changed fast, but the idea is the same: learn a person's look or voice, then generate new material.

  1. Learn a likeness

    Early face-swap deepfakes trained an autoencoder: a neural network that squeezes faces into a compact code and rebuilds them. Training on two people lets the system rebuild one person's expressions with the other's face. GANs and, more recently, diffusion models made the results sharper.

  2. Voices too

    Voice cloning can imitate a person's voice from a short recording. Scam calls that sound like a relative in trouble use this.

  3. Detection is an arms race

    Detection tools look for statistical traces of generation, but each new method leaves fewer. MIT's Detect Fakes project lists cues people can check: skin too smooth or too wrinkled, odd shadows around the eyes, glare on glasses, blinking too much or too little, and unnatural lip movement.

  4. Labels and provenance

    Another approach marks real and AI-made content at the source. The C2PA standard adds Content Credentials, a tamper-evident record of how a file was made and edited. Some AI tools add invisible watermarks.

Origin of the word
2017, from the username of a Reddit account that posted face-swapped videos; widely reported from December 2017
India's rule
IT Rules amended in February 2026: platforms must prominently label AI-generated content, and some takedown orders must be acted on within three hours
If you are targeted
In India, report at cybercrime.gov.in or call 1930. Save the link and screenshots first
Not all are harmful
Film de-ageing, dubbing into other languages and satire with clear labels use the same methods

We explain how deepfakes work so they are easier to recognise. We do not give instructions for making them.

Questions people ask

How can you tell if a video is a deepfake?

Watch the face edges, blinking, teeth and lips against the sound, and lighting that does not match the scene. Then check outside the video: is a trusted news source reporting it? Is it from the person's official account? Good fakes can pass the visual check, so the source check matters most.

Are deepfakes illegal in India?

Making a deepfake is not automatically illegal, but using one to cheat, impersonate, defame or share intimate images without consent is an offence under existing laws, including the IT Act. Since February 2026, platforms must label AI-generated content and remove unlawful material faster.

What should you do if someone makes a deepfake of you?

Save evidence: links, screenshots, dates. Report the post on the platform. In India, file a complaint at cybercrime.gov.in, or call the 1930 helpline and then file online. Tell people close to you so they are not fooled.

Can AI detect deepfakes?

Sometimes. Detection tools can catch many fakes, but new generation methods keep slipping past them, and false alarms happen. Treat a detector's result as one clue, alongside checking where the clip came from.

What is a voice deepfake scam?

A call or voice note that sounds like a family member or boss, asking urgently for money or a code. The voice is cloned with AI. Hang up and call the person back on a number you already know. A family code word helps.

Further reading

Prepared by Mundhra Labs. Reviewed by Aditya Mundhra. Checked October 2026.