Securing the Use of AI

What is AI voice cloning?

AI voice cloning is the use of artificial intelligence to recreate a specific person's voice from a few seconds of audio and then make it “say” anything, either as text turned into speech or live on a call. Criminals use the cloned voice of an executive or a family member to authorize payments and pry out information. It defeats the oldest proof of identity there is: recognizing who is speaking.

Zamak TechnologiesUpdated on July 12, 2026

How AI voice cloning works

Cloning a voice no longer takes a lab or hours of recording. With AI tools within reach of anyone, the scam comes together in four steps, and the scariest part is how little material it needs.

Collect a voice sample

A few seconds of public audio are enough: a social media video, a talk or webinar, a voicemail greeting, an audio message, even a short “wrong number” call made only to record the target speaking.

Train the voice model

An AI tool analyzes the sample and learns the timbre, accent and rhythm of the speech. In little time, it generates new sentences the person never said, with high similarity.

Write the pretext and choose the delivery

The criminal decides the script (the urgent transfer, the “emergency”, the request for a code) and how to deliver it: a pre-recorded audio, a voicemail, or real-time voice conversion, answering live as if they were the person.

Call under pressure

The victim gets a call from a voice they know and trust, wrapped in urgency and secrecy. By the time instinct says “it's really them”, the rational defense has already been bypassed.

Source: N-able Cyber Encyclopedia (vishing and AI voice cloning) and public reports on AI voice scams.

How to recognize a cloned-voice call

A call or an audio message with an urgent, secret request for a transfer, a code or a piece of data, outside the company's normal process.
The voice is familiar, but the number is unknown, or the topic and timing do not fit that person.
The caller resists hanging up to be called back on a known number, or moving to a video call with a live gesture.
Small oddities in the audio: flat emotion, pauses in the wrong places, breathing that does not match the speech, a background that is too silent or a quality that is too clean.
The push to act alone: “don't tell anyone” and “it can't wait”, precisely to keep you from confirming.

The most common forms of cloned-voice fraud

Executive fraud (the boss's voice) The cloned voice of a CEO or CFO calls or leaves an audio authorizing an urgent, confidential transfer. It is CEO fraud taken out of email and moved to voice, and far more convincing.
Family emergency (the “relative scam”) A voice identical to a child, a grandchild or a spouse calls in a panic, claiming an accident or an arrest and needing money right away. The target is emotion, not logic.
Voicemail and recorded audio Instead of a live call, the scam arrives as an audio message or a voicemail, which the victim listens to and forwards without suspicion.
Breaking voice biometrics Systems that use the voice as a password (“my voice is my password”) can be fooled by a good clone. That is why the voice should never be the only authentication factor.
Combined with other channels An email or a text sets the stage and asks the victim to “confirm by phone”; on the other end, the cloned voice closes the scam. Switching channels gives a false sense of legitimacy.

Why voice cloning is a real threat to the business

1 in 4

people, in a seven-country survey, have faced an AI voice scam or know someone who has (McAfee, 2023)

77%

of AI voice scam victims lost money (McAfee, 2023)

+148%

rise in impersonation scams over twelve months, with criminals using synthetic voices and agents (Identity Theft Resource Center, 2025)

Voice cloning is dangerous because it attacks a shortcut everyone trusts: the voice of someone we know has always been proof enough of identity. AI ended that certainty. In a seven-country survey, one in four people had already faced an AI voice scam or knew someone who had, and 77% of victims lost money (McAfee, 2023). Impersonation scams already cost consumers billions a year, according to consumer-protection authorities, and are growing fast: in a single twelve-month span, reports rose 148%, driven by synthetic voices (Identity Theft Resource Center, 2025). In the corporate world the script is simple: the cloned voice of an executive asks for an urgent, confidential transfer, and an employee under pressure, hearing the boss “in person”, pays. What makes the scam efficient is not the technology alone, it is that most people do not trust that they could tell a cloned voice from a real one. That is why the defense cannot depend on the ear of whoever answers the phone.

How to protect against voice cloning

None of these defenses depend on a sharp ear. They all start from a simple principle: treat the voice as a claim to verify, never as an identity already proven.

Confirm on a known channel, every timeFor any urgent request for money or data made by voice, hang up and call back the number you already have, not the one that called you. An independent second channel almost always breaks the scam.
Agree on a safe wordSet a secret word or question with your family and your finance team for urgent requests. A cloned voice imitates the timbre, but it does not know the agreed code.
Treat urgency and secrecy as a red flagEvery voice scam needs you to act fast and confirm with no one. Flip the rule: the more urgent and secret the request, the more it demands verification, not less.
Do not use the voice as a single passwordIf any system authenticates by voice, pair it with a second factor (MFA). Voice biometrics alone have stopped being reliable proof.
Reduce exposed audio and prepare the teamLimit what your voice leaves public where you can and, above all, train people: a team that knows the scam and has a verification process is the defense that works best.

In practice

For your entire life, recognizing a voice was proof of who was on the line. AI ended that rule: the voice of your director or your child is now something a scammer rents for a few dollars. The defense is not to listen more closely, it is to decide, before the phone ever rings, that no voice authorizes anything on its own.

How Zamak helps against voice cloning

Against voice cloning, Zamak Technologies starts from one principle: no voice, however familiar, authorizes a transfer or the release of data on its own. Within the managed cybersecurity of the Zamak Method, and alongside your team, Zamak helps install the habit of verifying on a second channel, trains whoever answers the phone to recognize the urgent, secret request, and keeps the identity and monitoring layers that catch unauthorized access when the scam gets through. Start by seeing where AI already exposes your company, with the AI exposure assessment.

Frequently asked questions about AI voice cloning

How much audio does it take to clone a voice?

Very little. Current tools reproduce a voice with high similarity from just a few seconds of audio, the kind anyone leaves public in a video, a talk or a voicemail greeting. Nothing has to be hacked: the material is usually already available.

What is the difference between a deepfake and voice cloning?

Voice cloning is the audio form of a deepfake: it recreates only the voice. Deepfake is the broader term, which also includes fake video and images, such as a video call with an executive's face. In phone scams, voice cloning is the central piece.

How do I know if a voice on a call is cloned?

Do not trust the ear alone, because a good clone fools it. Be suspicious of the whole picture: an urgent, secret request, an odd number or time, resistance to being called back, and small audio flaws such as flat emotion or pauses out of place. The only safe proof is to verify on another channel.

Can I trust voice biometrics (“my voice is my password”)?

Not as a single factor. The voice has become something AI can reproduce, so a system that authenticates by voice alone can be fooled by a clone. Voice biometrics are only safe when paired with a second authentication factor (MFA).

Does voice cloning threaten my personal life too?

Yes. One of the fastest-growing scams uses the cloned voice of a relative in a panic, claiming an accident or an arrest and needing money now. Agreeing on a safe word with your family and always calling back the known number protects the home as much as the company.

How does a company protect itself against voice cloning?

With process and preparation, not technology alone. Establish that every urgent voice request is confirmed on a second channel, agree on safe words, do not use the voice as a single password, and train whoever answers the phone. It is the same logic as defending against social engineering: what disarms the scam is a prepared team.