Can you tell if a voice or video call is an AI deepfake?

Increasingly, no — not reliably, and not in the moment. Real-time voice clones can be built from seconds of audio, and live video deepfakes now run on consumer hardware. The old visual tells (ask them to turn their head, wave a hand) are eroding fast. The only defense that doesn't expire is one that never depends on detection: a pre-agreed protocol — a safe word, a callback, a money rule — that works even if the fake is flawless.

What is a family safe word and how do I use it?

A safe word is a phrase every family member knows in advance and no stranger could guess. If you get an urgent call — a relative in trouble, a boss demanding a transfer — asking for money or secrecy, you ask for the safe word. A real person knows it instantly; an AI clone of their voice does not. Agree one at your next family gathering, keep it off social media, and use it any time a call pressures you toward money or silence.

What should I do if I get a scam call that sounds exactly like a family member?

Do not act on that call. Hang up and call the person back on the number you already have saved for them — the scammer controls the call they placed but cannot answer the number you dial. If you can't reach them, call another family member to verify. Never move money, buy gift cards, or send crypto on the strength of a single urgent call, no matter how real the voice sounds.

Toolkit · Family AI safety

The Family Safe-Word & Callback Card (2026)

Published July 4, 2026 · Vita Indarra

Short answer: AI can now clone a familiar voice from seconds of audio and fake a live video call convincingly. You cannot reliably spot it in the moment — so the winning move is to stop trying, and use a plan agreed in advance instead. Below is a one-card family protocol built on four rules that work even if the fake is perfect, because not one of them depends on you detecting it. Screenshot the card. Send it to your family. Agree your safe word this week.

The card · screenshot this

If a call or message pressures you toward money or secrecy:

Safe word. Ask for the family word only you know. A real relative knows it. A clone doesn't.
Call back. Hang up. Call them on the number you already have saved — never the number that called you.
Money rule. No money moves on one call or message. Gift cards, crypto, or a wire "right now, don't tell anyone" = scam, not emergency.
Two people. Any large or unusual payment needs a second person to confirm, separately.

The fake can be flawless. None of these four ask you to notice.

Why "just spot the fake" already failed

For a while, the advice was to look for tells — ask them to turn their head, watch for a glitch, listen for a flat note. That advice is expiring. Real-time voice cloning needs only seconds of your relative's audio (a voicemail, a video, a podcast), and live video deepfakes now run on ordinary hardware — in one 2024 case, a finance worker joined a video meeting where every "colleague" was a deepfake and wired about US$25 million. If your defense is detection, you are in an arms race you will eventually lose. The four rules above win a different way: they never require you to tell real from fake at all.

The four rules, a little deeper

The safe word is a shared secret — a phrase every family member knows and no stranger could find. Keep it off social media, and ask for it any time a call turns urgent about money or secrecy. The callback rule works because a scammer controls the call they placed but cannot answer the number you dial from your own contacts. The money rule targets the real signature of these scams — urgency, secrecy, and an untraceable payment method stacked together — none of which a genuine emergency actually requires. And dual approval is the same principle banks and finance teams use: one compromised channel should never be able to move money alone.

Want the "why" behind each threat? We keep plain-English explainers on voice-cloning scams, deepfake video calls, and AI romance scams. This page is the one-card summary they all point to.

Frequently asked

What if I don't have a safe word set up yet and get a call now?

Use the callback rule — hang up and dial the person on your saved number. It needs no prior setup and defeats the same attack. Then agree a safe word so next time is even faster.

Isn't asking a panicking relative for a password cold?

A real relative will understand in two seconds; the ask costs nothing. The scam depends entirely on rushing you past exactly this pause — which is why the pause is the defense.

Do these rules work for scam texts and emails too?

Yes. Callback, money rule, and dual approval are channel-agnostic: verify out of band, never move money on a single message, and require a second person for anything large.

Go deeper

The field guide behind this card

This card is the pocket version of Don't Trust the Robot — a plain-English guide to using AI without getting fooled, and spotting it when it's pointed at you. Voice clones, deepfakes, romance and investment cons, and the everyday habits that keep your family and money safe. Written for non-technical readers. Live on Amazon.

Don't Trust the Robot · $4.99 Which book should I read first?

← More field notes