Voice cloning scams and the one rule that defeats them

A relative calls, distressed. The voice is right — the pitch, the accent, the way they say your name. They need money now, and they cannot explain why on the phone.

This scam is old. What changed is that the voice is no longer an actor doing an approximation.

Why detection is the wrong goal

Advice that asks you to listen for artefacts is advice with an expiry date. Cloning has improved every year, and the operational reality is worse than the technology:

• You are hearing it under stress, which is the point of the story.
• You are hearing it over a phone line, which strips exactly the detail you would need.
• You are motivated to believe it, because the alternative is that someone you love is in trouble and you are wasting time.

Any defence that requires you to be perceptive at the worst moment of your week is not a defence.

The rule that works

Hang up and call back on a number you already have.

Not a number they give you. Not a callback to the number that appeared on your screen — caller ID is not verified and can be set by the caller. The number in your contacts, or one you look up independently.

If the call is genuine, you will reach them in thirty seconds and the delay costs nothing. If it is not, the scam ends there, because the attacker does not control the real person's phone.

This single rule defeats voice cloning, caller ID spoofing, and the entire family of impersonation calls at once — without requiring you to judge anything.

When calling back is not possible

The stories are built to close that door: my phone is broken, I'm being held, don't call anyone. So agree in advance:

A family password. A word or phrase, agreed out loud, never written in a message, never stored in a phone. Asked and answered before any urgent request is acted on. Choose something no one could research — not a pet's name, not a street you lived on.

Tell children, and tell older relatives, who are targeted disproportionately and often live alone with no one to sanity-check a story.

The tells that still hold

Not in the voice — in the structure. These do not improve with the technology:

• The call arrived; you did not place it.
• It cannot wait. Every real emergency survives a five-minute verification. Bail, fines, fees and transfers all still exist in an hour.
• The payment method is irreversible — transfer, cryptocurrency, gift cards, cash to a courier. This is the request, not a detail of it.
• You are told not to tell anyone. Secrecy is not part of any legitimate process, and it is the request most worth refusing on its own.
• The story explains why you cannot verify. That explanation is doing more work than the emergency.

Nothing in that list depends on how good the voice was. That is why it is the list to memorise.

Reduce what is available to clone

A few seconds of clear speech is enough, and most of us publish more than that:

• Voicemail greetings in your own voice — many networks offer a synthetic alternative.
• Public video and audio — talks, podcasts, social posts. Not a reason to stop, just a reason to know.
• Children's accounts, which are often public by default and full of speech.

You cannot remove all of it, and that is fine — the callback rule does not depend on the attacker lacking material.

Businesses are targeted the same way

The corporate version has an executive calling finance about an urgent confidential payment, sometimes on a video call with a synthetic face as well. The defence is the same shape and needs to be written down:

1. Payments above a threshold require a second approver, always, with no exception for urgency.
2. Verification happens on a known internal channel, never on the one the request arrived through.
3. "Confidential, do not discuss with the team" is escalated, not obeyed.

Anyone who cannot follow those rules because of who is asking is exactly who the scam is impersonating.

What is voice cloning?

Voice cloning is the synthesis of speech in a specific person's voice from a short sample of their real speech. Seconds of clear audio are enough, and voicemail greetings, videos, podcasts and social posts all supply it — which is why limiting your exposure is not a workable defence.

Why is detection the wrong approach?

Only the last 2 rows are defences. The first 4 ask you to be perceptive at the worst moment of your week, which is not a strategy.

What should families agree in advance?

• A password, spoken aloud, never written in a message or stored on a phone.
• Something unresearchable — not a pet, a birthday, or a former address.
• Told to children and older relatives, who are targeted disproportionately.
• A rule that no money moves without it, whatever the story.

The FTC's alert on AI-enhanced family emergency scams describes the pattern; the password is what makes it survivable. See phishing, impersonation and fraud prevention.

The business version

The corporate form has an executive calling finance about an urgent confidential payment, sometimes over video with a synthetic face. The defence is the same shape and has to be written down rather than assumed:

1. Payments above a threshold need a second approver, with no exception for urgency.
2. Verification happens on a known internal channel, never the one the request arrived on.
3. "Do not discuss this with the team" is escalated, not obeyed.

Voice cloning makes the first two non-negotiable. Anyone who cannot follow them because of who is asking is precisely who is being impersonated.

What to do if money has already moved

Voice cloning cases move fast because the story is built around urgency, so the response has to be faster:

• Call your bank on the number on your card and say the words "authorised push payment fraud" if you sent a transfer.
• Report within hours, not days. Many countries now have reimbursement rules that depend on prompt reporting.
• Warn the person who was impersonated, because their voice sample came from somewhere and they may be targeted again.
• Tell the rest of the family the same day, since the same operation frequently calls two relatives in sequence.

Reducing what is available to clone

You cannot remove all of it, and the callback rule does not depend on you trying. Still, a few sources are worth knowing about:

• Voicemail greetings in your own voice — most networks offer a synthetic alternative.
• Public talks, podcasts and videos, which are the richest samples available.
• Children's accounts, often public by default and full of speech.
• Voice notes in group chats that pass through many hands.

None of this is a reason to withdraw from public life. It is a reason to know that voice cloning needs no cooperation from you, and therefore that the defence has to sit somewhere the attacker cannot reach — which is the number you dial back on.