A video of a politician saying something they never said. A voicemail from your “boss” asking for an urgent bank transfer. A familiar face in a video call that isn’t quite right. Deepfake technology — AI-generated synthetic media of real people — has moved from research curiosity to everyday threat. Understanding what it is and how to spot it is now basic digital self-defence.

What Is Deepfake Technology?

A deepfake uses AI (typically a type of neural network called a GAN, or more recently diffusion models) trained on images and video of a real person to generate new footage of them saying or doing things they never did. The technology has two main forms:

• Face-swap deepfakes: replacing one person’s face with another’s in existing video. The most common type — used for everything from film effects to non-consensual imagery.
• Full synthesis: generating entirely new video (and increasingly audio) of a person from scratch, driven by a script or another performer’s movements.
• Voice cloning: the audio cousin — AI trained on a few minutes of someone’s voice can generate new speech in it. This powers many phone scams.

What changed recently is accessibility: tools that once needed expertise and computing power now run as simple apps. That’s why deepfakes went from rare to routine in just a couple of years.

The Real Harms

• Non-consensual intimate imagery. The largest category of deepfake abuse by far — overwhelmingly targeting women and girls, including schoolchildren. Creating or sharing these is illegal in the UK under the Online Safety Act provisions.
• Financial fraud. Voice-cloned “bosses” authorising transfers, “family members” in distress asking for money. UK businesses and families have lost significant sums.
• Political disinformation. Fake videos of candidates or officials, timed to elections or crises.
• Reputational attacks. Fabricated compromising footage used for blackmail or harassment.

How to Spot a Deepfake: Video Tells

1. Unnatural blinking or eye movement. Early deepfakes rarely blinked; modern ones are better, but gaze can still feel “off.”
2. Face-body mismatches. Skin tone differences between face and neck, blurry edges around the jawline, inconsistent lighting on the face versus the scene.
3. Weird mouth movements. Lip-sync that’s slightly out, teeth that look painted on, or mouths that don’t quite match the audio.
4. Lack of natural imperfection. Overly smooth skin, no natural head micro-movements, robotic stillness.
5. Audio-visual mismatch. Voice that doesn’t quite sync, or emotional tone that doesn’t match facial expressions.
6. Context scepticism. Does the source make sense? Would this person really say this, here, now? Extraordinary claims need ordinary verification.

Spotting voice-clone scams

• Establish a family code word for emergency money requests — low-tech, highly effective.
• Call back on a known number. Never act on an urgent voice request without verifying through a separate channel.
• Listen for oddities: flat emotion, strange pauses, mispronounced names the real person would know.

For still images, see our companion guide: how to spot AI-generated fake images.

What to Do If You’re Targeted

• Don’t engage with blackmailers. Paying never ends it.
• Preserve evidence — screenshots, URLs, messages.
• Report it: to the platform, to the police (non-consensual intimate deepfakes are criminal offences), and to the Internet Watch Foundation for image-based abuse.
• Tell someone. Shame keeps victims silent; the support services (including the Revenge Porn Helpline) deal with this daily and without judgement.

Protecting Yourself Proactively

• Limit public video/audio of yourself where practical (hard in the social media age, but be aware).
• Use privacy settings on social accounts.
• Set up that family code word now, before you need it.
• Teach kids about deepfakes early — see AI risks for children.

Deepfakes and UK Law: What You Should Know

• Non-consensual intimate deepfakes: creating or sharing sexually explicit deepfakes without consent is a criminal offence in the UK, carrying serious penalties. This includes “just sharing” something someone sent you.
• Fraud: using deepfakes for financial fraud is prosecuted under fraud legislation — the technology doesn’t create a legal grey area, just a new method for old crimes.
• Online Safety Act: places duties on platforms to address harmful synthetic content, including deepfake abuse. Ofcom enforces these duties.
• Defamation: deepfakes that damage someone’s reputation can ground civil claims — the depicted person doesn’t need to prove who made it to seek platform takedowns.
• Election law: materially misleading deepfakes around elections face increasing scrutiny — check the Electoral Commission’s guidance during election periods.

The law is clearer than many realise: existing offences cover most deepfake harms — the challenge is enforcement and identification, not legislation.

Deepfakes in the Workplace

Businesses are increasingly targeted:

• CEO fraud: voice-cloned executives instructing finance staff to make urgent transfers. Defence: verification procedures for all payment instructions, no exceptions — see banking fraud detection.
• Fake video meetings: entirely fabricated video calls with “colleagues” — documented in several major fraud cases. Defence: multi-channel verification for unusual requests.
• Recruitment fraud: deepfaked candidates in remote interviews (someone else doing the interview, another doing the job). HR teams need verification procedures.
• Policies to implement: a code word for financial instructions, mandatory call-backs on known numbers, and staff training. Our guide to free AI tools for small business covers the legitimate side; this is the defensive side.

The 30-second business rule

Any urgent, unusual financial instruction received by voice or video gets verified through a second, independent channel before action. No exceptions, no matter how senior the apparent source or how urgent the tone. This single rule defeats nearly all deepfake fraud.

The single most effective deepfake defence remains the least technical: slow down. Verify through a second channel, check the source, and never act on urgency alone. Technology created this problem; calm human judgement solves most of it.

Practical Takeaway

Deepfake technology uses AI to fabricate convincing video and audio of real people — and it’s now easy enough for anyone to use. Your defences are scepticism (verify extraordinary media before believing or sharing), verification habits (call back on known numbers, check primary sources), and knowing where to report abuse. The technology will keep improving; the tells will fade. What won’t fade is the value of pausing before you believe your eyes.