Type of cyber crime
AI deepfake & voice-clone scams
Cloned voices of relatives, deepfake videos of officials and celebrities, and AI-generated profiles used to make every other scam more convincing.
Generative AI did not invent new scams; it removed the tells. A few seconds of audio from a social-media reel is enough to clone a voice that calls a parent about an accident. Deepfake videos of finance ministers and industrialists front investment ads. "Officers" on digital-arrest calls are sometimes synthetic. Detection is getting harder; verification habits still work.
How it usually looks
- A call in your son's voice, crying, asking for money after a "police case".
- A video ad of a well-known investor endorsing a trading app.
- A video-call "bank manager" whose face flickers when they turn their head.
Law that applies
| Provision | What it covers | Punishment |
|---|---|---|
| IT Act 2000 s.66D | Cheating by personation using a computer resource | Up to 3 years and fine |
| IT Act 2000 s.66E / s.67 | Privacy violation; obscene deepfakes | Up to 3 years |
| BNS 2023 s.318 / s.319 / s.356 | Cheating; personation; defamation | Up to 7 years for cheating |
| IT Rules 2021 (as amended) | Platform duties to remove deepfakes on notice | Loss of safe harbour |
Sections are quoted for orientation; the police decide the final charge sheet. Verify text at indiacode.nic.in.
Where to report
- Verify with a call back on the known number or an agreed family "safe word" before sending anything.
- If money moved, 1930 and cybercrime.gov.in.
- Report deepfake content to the platform as impersonation; public figures can also approach the courts for takedown orders.
How to prevent it
- A family safe word for emergencies.
- Treat urgency plus a request for money as the signal, regardless of how real the voice is.
- Keep voice-heavy public content to a minimum for children and elders.
Frequently asked questions
Can a deepfake be detected?
Sometimes, by artefacts around the mouth, eyes and ears, or by asking the caller to perform an unexpected action. But detection is unreliable and getting worse. Verification through a second channel is the dependable defence.