Cybernion
← Insights

Security

Deepfake and Voice-Clone Fraud Explained

30 Aug 2026 · Updated 30 Aug 2026

Deepfake and voice-clone fraud uses AI to impersonate a trusted executive in a call or video meeting, turning business email compromise into something far more convincing. In one widely reported case, staff at engineering firm Arup transferred around 25 million US dollars after a video call where every other participant was an AI deepfake. The defence is not detection technology but process: verify every payment instruction out-of-band, through a known, separate channel.

The scam where the “CEO” emails asking for an urgent payment has evolved. Now the CEO can call, or appear on a video meeting, and it can be entirely fake. This is one of the fastest-moving fraud threats, and the defence is more about process than technology.

What is deepfake and voice-clone fraud?

Deepfake and voice-clone fraud uses AI-generated video or synthetic voice to impersonate a real person, almost always a senior executive, to trick an employee into authorising a payment or handing over sensitive information. It is the next stage of business email compromise and CEO fraud: the same social engineering goal, but delivered through a fake voice call or video meeting rather than a spoofed email.

What makes it a distinct threat is that it attacks a deeper instinct. People have learned to be a little sceptical of emails, but a phone call in the boss’s voice, or a video call showing the boss’s face, bypasses that scepticism, because we are wired to trust a familiar face and voice. Voice cloning can now reproduce a person’s voice from a short sample, and video deepfakes can put a convincing likeness on a live call. As the technology has become cheaper and more accessible, this has moved from a theoretical risk to an active fraud method that Australian businesses need to plan for.

How does a deepfake fraud attack work?

The pattern is consistent. Attackers gather publicly available footage and audio of the target executive, from conference talks, media, social media and company videos, and use AI to reconstruct their face and voice. They then impersonate the executive, typically to instruct a finance employee to make an urgent payment, often reinforced by an initial spear-phishing message to set up the contact.

The most striking documented example is the case of engineering firm Arup. As widely reported, including by the World Economic Forum, a finance employee in the Hong Kong office joined a video call in which the chief financial officer and other colleagues were all AI-generated deepfakes, and, convinced it was genuine, made a series of transfers totalling around 25 million US dollars. The detail that matters is the use of multiple synthetic participants: it was not one fake voice but a whole fake meeting, which created powerful social proof. The attack combined public source material, convincing real-time deepfakes, and the authority and urgency of a senior instruction, which is the template these attacks follow.

Why is deepfake fraud so effective?

Deepfake fraud is effective because it stacks several psychological levers that each defeat normal caution. The first is trust in a familiar face and voice, which we grant almost automatically. The second is social proof, an attacker can populate a fake video call with several synthetic colleagues, so the victim sees a room full of familiar people apparently agreeing, which is enormously reassuring and hard to doubt in the moment.

On top of these, the attacks use the same authority and urgency that make email-based CEO fraud work: an urgent instruction from the apparent chief executive, framed as time-critical and confidential, pressures the employee to act before they think to verify. The combination overrides the scepticism someone might apply to a suspicious email. And the barrier to mounting these attacks has fallen sharply, convincing voice and video impersonation no longer requires a specialist or much time, which means more attackers can use the technique against more targets. The uncomfortable implication is that no organisation can rely on staff to spot a good deepfake, because the whole point is that it is convincing, which is why the defence has to be process, not perception.

How is deepfake and voice-clone fraud defended against?

The core defence is out-of-band verification, and it is simple to state: never act on a payment or sensitive instruction based on a call or video alone, always confirm through a separate, known channel. If the “CFO” calls asking for an urgent transfer, the employee calls the CFO back on their known number, or verifies through an agreed internal channel, before acting. Because the attack lives entirely within the fake call, stepping outside it to verify breaks it.

Around that core, layer the controls that stop payment fraud generally. Require dual control and step-up approvals for high value transfers, so no single person can move large sums alone. Build in cooling-off or review windows for large or unusual payments, which remove the urgency the attack depends on. Agree challenge phrases or code words for executives to use when authorising sensitive actions. Enforce strict payment authorisation workflows, especially for any change to bank or payment details. And train finance and executive staff specifically on this threat, so they expect it and know the verification step is mandatory, not optional. Deepfake detection tools exist and can help, but treat them as a supplement, not the primary control, because the attacker only has to beat the tool once, whereas a mandatory out-of-band verification step stops the attack regardless of how good the deepfake is. For an organisation wanting help hardening its payment and verification processes, Cybernion helps through advisory.

What should Australian businesses do?

Treat deepfake fraud as the next stage of the business email compromise and payment-redirection fraud that the ASD and Scamwatch already warn Australian businesses about. The good news is that the controls that stop email-based payment fraud, out-of-band verification, dual authorisation, strict change-of-details processes, also stop the deepfake version, so strengthening those is the highest-value action.

Concretely: put strict payment verification workflows in place, require out-of-band confirmation and dual authorisation for any change to payment details and any large or unusual transfer, and make those steps mandatory with no exceptions for “urgent” requests, because urgency is the attacker’s tool. Train finance teams and executives on the threat, and rehearse the response so it is second nature. For an Australian business, this connects to the broader defence against business email compromise and payment fraud, and to a security culture where verifying a request is normal rather than seen as distrust. The organisations that avoid these losses are the ones where verifying a payment instruction is simply how things are done, regardless of who appears to be asking.

Frequently asked questions

What is deepfake and voice-clone fraud?

Deepfake and voice-clone fraud uses AI-generated video or synthetic voice to impersonate a real person, usually a senior executive, to trick staff into authorising payments or handing over information. It is the evolution of business email compromise and CEO fraud, moving from spoofed emails to fake voice calls and even fake video meetings. Because it impersonates a trusted individual convincingly, it defeats the instinct to trust a familiar face or voice.

How does a deepfake fraud attack work?

Attackers gather publicly available footage and audio of the target executive, use AI to reconstruct their face and voice, and then impersonate them, often in a phone call or video meeting, to instruct an employee to make an urgent payment. In one widely reported case, a finance employee at engineering firm Arup joined a video call where the CFO and colleagues were all AI-generated deepfakes, and made transfers totalling around 25 million US dollars.

Why is deepfake fraud so effective?

It is effective because it exploits trust in a familiar face and voice, and adds social proof, an attacker can populate a fake video call with several synthetic colleagues so the victim feels reassured. Combined with urgency and authority, an urgent instruction from the apparent CEO, it overrides the normal caution someone might apply to an email. The technology has advanced to the point where a convincing impersonation no longer requires much time or skill.

How do you defend against deepfake and voice-clone fraud?

The core defence is out-of-band verification, never act on a payment or sensitive instruction based on a call or video alone, always confirm through a separate, known channel such as calling back a known number. Add dual control and step-up approvals for high value transfers, cooling-off periods, agreed challenge phrases for executives, and strong payment authorisation workflows. Staff training is essential, and deepfake detection tools are a supplement, not a primary control.

What should Australian businesses do about deepfake fraud?

Treat it as the next stage of business email compromise and payment fraud, which the ASD and Scamwatch already warn about. Put strict payment verification workflows in place, require out-of-band confirmation and dual authorisation for any change to payment details or any large transfer, train finance and executive staff on the threat, and rehearse the response. The controls that stop email-based payment fraud also stop the deepfake version, so strengthen those first.


Written by Gaurav Vikash, an ASD endorsed IRAP assessor and senior cyber security leader with 18 years of experience across Australia, the UK and Asia, including CISO and senior security leadership roles. He holds CISSP, CISA, CISM and CRISC and is an ISO 27001 Lead Auditor and ISO 42001 Lead Implementer, and speaks regularly at industry conferences.

Talk to us.

Sources:

  1. How a deepfake scam cost a company 25 million dollars, World Economic Forum, February 2025
  2. Engaging with Artificial Intelligence, Australian Signals Directorate, 2024
  3. Business email compromise, Australian Signals Directorate, 2026
  4. Scamwatch, National Anti-Scam Centre, 2026

Last updated: 30 August, 2026

Talk to us about your engagement

A scoped proposal within one business day.