All articles AI & Deepfakes

How to Spot AI Deepfakes

6 min read

Deepfakes used to be easy to spot — wonky teeth, weird eyes, robotic voices. Not anymore. In 2026, AI-generated video and audio is good enough to fool families, journalists and finance teams. KNOMI Cyber sees more deepfake-driven scams every quarter, and the defence is a small number of habits anyone can learn.

What to look for in video

Watch the edges: hairlines that ripple, jewellery that shimmers, glasses that flicker between frames. Watch lighting: many deepfakes use a single source and shadows that don't match the background. Watch the mouth: synthetic lip-sync still tends to over-articulate consonants and miss the small mouth movements between words.

What to listen for in audio

AI voice clones can be trained from 30 seconds of a podcast. The giveaways are increasingly subtle — flat emotional range, missing breath sounds, slightly mechanical pacing on numbers and names, and an unwillingness to be interrupted. If a 'family member' calls in distress and asks for money, hang up and call them back on a known number.

Deepfake red flags

  • Urgency combined with a request for money or codes
  • Inability to verify with a safe word
  • A 'live' video that looks like a single static head
  • Audio with no background noise at all
  • Refusal to do a simple visual gesture (turn head, wave)

What KNOMI does about it

KNOMI Cyber helps Australians preserve deepfake content as evidence (links, hashes and metadata matter), report it to eSafety where it involves intimate or threatening material, and build a containment plan when an executive, parent or public figure is being impersonated.

Frequently asked questions

Are deepfakes illegal in Australia?

Sexually explicit deepfakes are criminalised under federal law. Many other uses fall under defamation, fraud or eSafety's image-based abuse scheme.

Can KNOMI take down a deepfake?

KNOMI helps you build the takedown request and engage eSafety, the platform, and where needed, legal counsel. We don't operate the takedown ourselves.

How do families defend against voice clones?

Agree on a family safe word, call back on known numbers, and never act on urgent money requests from voice alone.