05
The voice on the phone is your daughter. Probably not.
A few seconds of audio is enough to clone someone. The defence is not technical and it costs nothing: agree a password with your family tonight.
5 min read
Voice cloning went from a research demonstration to a consumer product in a very short time. It now needs a small sample — seconds, not minutes — and the sample is usually already public, in a voice note, a video, a work presentation.
The resulting fraud is simple and it works because it targets something below reasoning.
The shape of the call
It arrives as an emergency, usually late. A familiar voice, distressed, brief. An accident, an arrest, a hospital, a phone taken away. Someone else comes on the line to explain what is needed. Money, now, and do not tell anyone.
Every element is doing a job:
- Distress shuts down evaluation. You are not assessing voice quality, you are reacting.
- Urgency removes the pause in which you would call back.
- Brevity limits exposure. The clone is not asked to hold a long conversation.
- Secrecy stops you from checking with the one person who would know.
Note how little of this depends on the audio being perfect.
Why the usual advice fails
Listen carefully to the voice. Recognition is precisely the faculty under attack. You will recognise it, because it is a copy of something you recognise.
Check the caller ID. Numbers are spoofed routinely.
They will not have enough audio to clone. They will.
What actually works
A family password. A word or short phrase, agreed in advance, out loud, never written in a message. If someone calls claiming to be family in trouble, you ask for it.
That is the whole defence. It costs one conversation, it is not defeated by better models, and it does not require you to be calm.
Choose something arbitrary and memorable. Not a pet, a birthday, a street or anything else discoverable.
Hang up and call back, on the number you already have. Not a number given to you on the call.
Ask something only they would know, with no online trace. Not “where did we go last summer” — a small, undocumented, shared detail.
The tells, for what they are worth
Cloned voices still tend to sit in unnatural silence, with a missing or mismatched room tone. Breathing is often wrong: absent, or placed evenly rather than before long sentences. Interruption is difficult — a clone struggles to be talked over and to resume mid-word.
Talk over them. Ask them to repeat something backwards. Ask a question that needs a specific answer, not an emotional one.
But treat all of that as secondary. The password is the defence. The tells are what you fall back on when you did not set one up.
Do it now
Message the people who might one day call you at two in the morning. Agree a word. It will take four minutes and it is the highest-return security measure available to a household.