You pick up the phone. The number is unknown, but the voice on the other end is unmistakable. It's your mother. She's crying. She tells you she's been in a car accident, her phone is broken, she's at the police station, she needs you to pay her train ticket home — €500, urgently. Can you give her your card number, just this once?
For a fraction of a second, the part of your brain that recognizes your mother's voice — a primitive, pre-rational part — says yes, that's her. The doubt comes later. But the scammer doesn't need much. A few seconds of doubt-free response is all they need.
Welcome to the AI voice-cloning family emergency scam — what France's national cybersecurity agency Cybermalveillance.gouv.fr now calls "hypertrucage" (deepfake) and which they identified in their 2024 annual report as a new and growing threat vector, alongside phishing, romance scams and sextortion, all "increasingly sophisticated due to criminal use of generative AI."
How little voice the scammer needs
Three seconds. That's it.
According to research summarized by Bitdefender, modern AI voice-cloning tools can produce a convincing fake of someone's voice from as little as 3 seconds of clean audio. That's:
- A single voice note your child posted on Instagram Stories
- The intro of a TikTok video where someone says their name
- A WhatsApp voicemail you forwarded once
- A YouTube comment-section greeting
Jerome Notin, director of Cybermalveillance.gouv.fr, confirmed on France Bleu that the samples can be "scraped from social media in seconds." There's no defense via "they could never get my voice" — if your relative has any public-facing social media at all, their voice is available.
The current generation of cloning models even transfers emotion: crying, panic, breathlessness all come across. The same model that can clone your mother's normal speaking voice can make her sob.
What the scam call sounds like
The scammers follow a tight script designed to bypass your judgment:
Step 1 — Emotional shock: The call opens with the cloned voice in distress. "Allô, maman ? C'est moi… il y a eu un accident…"
Step 2 — Plausible disaster: Car accident, lost phone, arrested, abroad, hospital. Something dramatic but mundane enough to be believable.
Step 3 — Loss of normal contact: "My phone is broken / lost / the police took it." This explains why you can't see their usual number.
Step 4 — A handoff to a "helper": A "police officer", "hospital administrator", "consulate agent", or "lawyer" takes the phone. Now the cloned voice is gone and the scammer can speak normally without giving themselves away.
Step 5 — Money request: Bail money, train ticket, hospital bill, lawyer fees, customs deposit. Anywhere from €200 to €5,000.
Step 6 — Urgency: "It has to happen in the next 30 minutes or [bad thing]." This is the part where you don't have time to call back.
A documented 2024 case reported by TF1 involved a man receiving a call mimicking his mother's voice, pressuring him for his bank card number under the pretext of urgently purchasing train tickets after a claimed lost wallet/phone. The verbatim quoted line: "Elle me demande si je peux lui donner mon numéro de carte bancaire."
Real cases
- France, 2024 (TF1 / France Bleu): The case above — a man receives a perfect-sounding voice call from "his mother" asking for his bank card number to buy train tickets. The voice was a clone built from her social-media posts.
- Belgium, 2024 (RTBF investigation): RTBF's investigation documents a Belgian woman who lost €410,000 through an AI-deepfaked video of a public figure — the video and voice combination made the recommendation undeniable to her. The journalists traced the operation back to its source.
- Belgium, ongoing (Federal Police): RTBF's separate alert covers an active warning from Belgian police about voice deepfakes specifically, with concrete instructions to relatives.
- France, official agency naming: Cybermalveillance.gouv.fr's 2024 report explicitly attributes the rising sophistication of phishing, romance scams, sextortion, and "audio/video/photo deepfakes" to "growing criminal use of generative AI" — the first time the national agency has named AI voice-cloning in its official threat landscape.
In Morocco and Algeria, the same scam adapts to local family structures: the cloned voice is often a child living abroad ("Maman, je suis bloqué à Paris/Marseille/Brussels…"), which makes the "different unfamiliar number" detail completely plausible and harder to detect.
Why the voice alone fools you
There's a reason your gut reacts before your brain does. The auditory cortex recognizes familiar voices using cues that we can't consciously articulate — micro-rhythms of breathing, vowel timbre, characteristic pauses. AI cloning models now reproduce these with frightening accuracy. You can't think your way through a clone in real time the way you can stare at a phishing email and spot the wrong logo.
The scammer also uses every trick to disable your reasoning:
- They open with strong emotion (crying, breathless, panicked) which overrides your skepticism
- They invent a story that explains the unfamiliar number (broken phone, foreign country, police custody)
- They hand off to an authority figure as soon as the voice has done its job
- They rush you so you can't call back
The single technique that defeats all of this isn't recognizing the fake. It's refusing to act on the original.
The one rule that immunizes you and your family
Hang up. Call back on the number you know.
That's it. There's no AI in the world that can reach you when you initiate a call to a number you saved long before the call.
Cybermalveillance.gouv.fr and Belgian Federal Police both recommend the same defense, phrased the same way by the French source: "Dans le doute, il est conseillé de raccrocher et d'appeler le vrai numéro." In English: when in doubt, hang up and call the real number.
If your relative really is in trouble, calling them back costs you 30 seconds. If it's a scam, that 30 seconds saves you €500–€5,000.
The family password — your secret weapon
Set up a family code word today. A word that:
- Has nothing to do with you (not a pet's name, a city, a sports team — too guessable)
- Was decided in person, not over text or call
- Every adult in the family knows
- You agree to require any time someone calls asking for money
When the "your mother" call comes in, ask: "What's the family code word?"
The scammer doesn't have it. The real relative does. End of scam.
This is the single most effective defense against voice-clone scams, and it costs nothing.
What to do if you receive the call
- Don't share any payment info. Not bank details, not card numbers, not crypto addresses, nothing.
- Ask for the family code word. If they hesitate or get angry — it's a scam.
- Hang up. Even if you think it's real but they refuse the code word — hang up.
- Call the real number. The one in your phonebook. Not the one that just called you.
- If you can't reach them, call another family member. Ask them to verify.
- Report the call. In France: Cybermalveillance.gouv.fr and dial 17 (police). In Belgium: safeonweb.be and dial 101. In Morocco: E-Blagh or nearest commissariat.
What to do if you already sent money
- Call your bank immediately. Same day matters. Time is the difference between recoverable and gone.
- File a criminal complaint. France: Pré-plainte en ligne. Belgium: federal police + report to CCB. Morocco: commissariat + E-Blagh.
- Warn the family. Especially the relative whose voice was cloned — they need to know their voice is "in the wild" and may be used again.
- Audit their social media. Lock down voice notes, public stories, public reels. Remove voicemails accessible publicly.
Reduce your family's exposure
- Set Instagram / TikTok / Snapchat to private, especially for voices of children and elderly relatives.
- Don't post voice notes publicly.
- Set your WhatsApp voicemail privacy to "My Contacts" only.
- Delete old public voice content (YouTube comments, podcast guest appearances, etc.).
- Once a year, audit what's publicly searchable. Search your name + "voice" on Google.
Don't trust ears. Trust the callback.
In 2026, your ears can be deceived. Your saved phone numbers can't. Always call back on the known number before doing anything financial. Tell your parents this. Tell your kids. Tell your grandparents. Set the code word today.
Doubt? Verify in 5 seconds.
If you receive a suspicious message (the call is often followed up by a WhatsApp text from the "new number"), forward the text to Digiscam on WhatsApp or paste it into our check box. AI verdict in seconds. Free, anonymous, EN/FR/AR.
Sources: Cybermalveillance.gouv.fr 2024 annual report (PDF) · France Bleu — Interview Jerome Notin on AI voice cloning · RTBF — €410,000 lost to AI deepfake · RTBF — Belgian police alert on voice deepfakes · Bitdefender — Voice cloning from 3 seconds of audio