Deepfake Voice Scams

Deepfake voice scams use AI voice-cloning technology to replicate the voice of a family member, friend, or company executive with startling accuracy. The technology requires only a few seconds of audio — readily harvested from a social media video, voicemail greeting, or podcast appearance — to train a model that can say anything the scammer types.

The resulting call sounds unmistakably like the person you love or work for. Vocal cadence, regional accent, and even characteristic filler words carry over. Older telephone infrastructure further compresses audio quality, making artefacts harder to detect. The result is that even people who consider themselves tech-savvy have been deceived.

In family-emergency variants, scammers impersonate a child, grandchild, or sibling who has supposedly been in an accident, arrested, or kidnapped abroad. In corporate variants — sometimes called CEO fraud with a voice layer — finance staff receive calls from an 'executive' or 'auditor' directing an urgent and confidential bank transfer. Both variants share the same psychological engine: panic combined with a trusted voice short-circuits rational scrutiny, making verification feel disloyal or time-wasting.

The preparation phase takes minutes. Scammers scrape publicly available audio from social media reels, TikTok videos, YouTube interviews, or podcast appearances and feed it into a commercially available voice-synthesis service. Most modern platforms can clone a voice from under thirty seconds of clean audio and generate unlimited new speech.

The call itself is short and designed to disorient. The 'relative' or 'executive' establishes urgency immediately — there has been an accident, an arrest, a confidential deal — and stresses that time is critical. The caller pressures you to keep the matter secret, which isolates you from anyone who might counsel caution. They will have details scraped from social media: the target's name, family relationships, recent travel, or professional context.

Payment is always immediate and in a form that is hard to reverse. Wire transfers, gift card numbers read over the phone, or a crypto wallet address are all common. In business settings, the scammer may follow up with an email confirming the 'executive's' instruction to add a veneer of documentation. Once the money moves, recovery is extremely difficult.

Afterwards, scammers sometimes call a second time impersonating a 'recovery agent' or 'police officer' to extract further payments or personal information from distressed victims.

Voice is one of the most emotionally loaded channels of human communication. We associate the sound of a person's voice with trust, intimacy, and identity in a way that text or even images do not replicate. When you hear what sounds exactly like your child calling in distress, the emotional response — love, fear, urgency — is involuntary and powerful.

AI voice cloning removes the one safeguard people previously relied on: 'I would know my own family's voice.' That mental model has not yet updated to account for how accessible and how convincing modern synthesis is. The technology is now consumer-grade and requires no specialist knowledge.

Scammers layer additional psychological pressure on top: urgency (act now or it gets worse), secrecy (don't tell anyone), and authority (a lawyer or officer is on the line). These combine to suppress the analytical thinking that might trigger verification behaviour. The familiar voice provides false reassurance while the pressure mechanisms prevent the pause needed to act on doubt.

An older adult receives a distressed call from someone speaking in what sounds exactly like their adult child's voice. The 'child' says they have been in a minor car accident abroad and been detained; they need bail money wired immediately and beg the parent not to call other family members. A second voice — a 'lawyer' or 'consular officer' — takes the phone and provides wire-transfer instructions. The parent complies, believing they are helping their child. When they later reach their actual child, who is completely unaware, both parties realise what happened.

Mum, it's me — I've been in an accident and need you to send [amount] right now, please don't tell Dad.

This is [name]'s lawyer. Your [relative] needs [amount] for bail — we need it wired to [account] within the hour.

[Executive name]: Process this wire to [account] immediately — it's confidential, do not discuss it with anyone.

I've lost my wallet and passport in [city]. Can you send [amount] in gift cards and give me the codes? I'll pay you back.

Don't hang up — if you put the phone down they will charge [name] with a more serious offence.

It's [friend]. I'm in a really bad situation. I need [amount] urgently. I'll explain everything when I get back.

The single most important protection is a pre-agreed family safe word or passphrase — a short, memorable word that any family member can ask for in a suspected emergency. If the caller cannot supply it, treat the call as fraudulent regardless of how convincing the voice sounds.

After receiving any urgent call asking for money, hang up and call the person back directly on the number saved in your own contacts — not a number the caller gives you. If you cannot reach them, call another family member or colleague who can physically locate them.

For business calls directing transfers, use your organisation's established callback or dual-authorisation procedure. No legitimate executive will object to a standard verification step. Treat any instruction to bypass controls as a major red flag.

Consider keeping a private note of your family safe word in a password manager or secure location. Review it periodically so everyone remembers it without prompting. A well-established safe word is useless if family members have forgotten it when they need it most.