Article

Deepfake CEO Scams: Voice Cloning and BEC

·By Mathew Chewing

Scammers can now clone your boss's voice and face. How deepfake CEO scams work, the real cases behind the warnings, and the verification rules that stop the payment going out.

The “Deepfake CEO” Scam: Why Voice Cloning Is the New Business Email Compromise (BEC)

In January 2024, a finance employee at the Hong Kong office of Arup, the global engineering firm, got a message from the company's UK-based chief financial officer about a confidential transaction. He was suspicious. So he joined a video call to check, and there was the CFO, along with several colleagues he recognised, all looking and sounding exactly as they should.

Reassured, he made 15 transfers totalling HK$200 million, about US$25 million. Every other person on that call was a deepfake.

That case is the reason this article exists. Business email compromise used to rely on a convincing email. Now it can come with a voice, a face and a whole meeting full of fake colleagues.

Business email compromise, version two

Classic business email compromise (BEC) works like this: an attacker either breaks into a real email account or creates a lookalike, then sends a request that seems routine. Pay this invoice. Update this supplier's bank details. Buy some gift cards for a client. It's already one of the most expensive cybercrimes for Australian businesses. In the ASD's latest annual threat report, business email compromise fraud with a financial loss made up 15 per cent of cybercrime reports from businesses, and the average self-reported cost of cybercrime per report was about $56,600 for small businesses and $97,200 for medium ones.

Over the years, businesses got better at spotting the email. Staff learned to check sender addresses and question odd grammar. So the attackers added a channel that people don't check: the human voice.

How voice cloning actually works

A voice-cloning model learns the features that make a voice recognisable, such as pitch, accent, rhythm and the way certain sounds are formed, from sample recordings. Then it can say any text in that voice. Modern tools need surprisingly little audio to produce something passable, and executives give it away constantly: conference talks, podcast appearances, webinars, LinkedIn videos, even a voicemail greeting.

The attacker types a script. The model speaks it. Some tools now do this in close to real time, so the attacker can hold a conversation rather than play a recording. Video deepfakes take more effort, but as Arup showed, they're well within reach of an organised group.

Which brings us to an uncomfortable bit. We built software that can make anyone say anything in their own voice, and we made it cheap and easy to use. It's genuinely impressive engineering. And it is, uh... exactly the kind of impressive that should make every finance team a little nervous.

Why it works on smart people

Nobody falls for these because they're foolish. They work because they hit three buttons at once:

  • Authority. Most people don't push back on the CEO. Questioning the boss feels riskier than doing what you're told.
  • Urgency. The deal closes today. The supplier will walk. It has to happen before the weekend.
  • Secrecy. "This acquisition is confidential, don't discuss it with anyone." Which neatly stops you checking with the one person who'd spot the fraud.

Add a voice you recognise and the email checks you were trained on become irrelevant. You're not reading headers; you're talking to your boss.

The case that went the other way

In July 2024 an executive at Ferrari got WhatsApp messages, then a call, from someone who sounded exactly like the CEO, Benedetto Vigna, accent and all, talking about a confidential acquisition. Something felt slightly off, so the executive asked a simple question: what was the title of the book Vigna had recommended to him a few days earlier? The caller hung up.

That's the lesson in miniature. The executive didn't need deepfake detection software. He needed one piece of information the attacker couldn't have, and the confidence to ask for it.

Can you hear the difference?

Sometimes. Reported tells include slightly mechanical rhythm, odd breathing, strange pauses before answers and background noise that doesn't fit. In the Ferrari case the executive noticed the cadence was a bit robotic.

But don't build your defence on your ears. The technology improves every few months, phone lines compress audio and hide artefacts, and a stressed person on a call isn't doing careful audio analysis. Assume you can't tell, and design your processes so you don't need to.

The controls that actually stop it

The good news is that the defence is process, not technology, and it works against every version of this scam: email, voice or video.

1. Call back on a number you already have

Any request to move money, change bank details or share sensitive data gets verified by contacting the person through a channel you choose: their number from the staff directory or your phone's contacts, never the number that called you or one supplied in the message.

2. Two people for every payment

Set up your banking so payments above a threshold, and all new payees, need a second approver. An attacker now has to fool two people independently, which is much harder.

3. A verification phrase for urgent requests

Some businesses agree a code word that executives use for genuine urgent requests, changed regularly and never written in email. The Ferrari question did the same job on the fly.

4. Permission to say no

This is the important one. Leadership needs to tell staff, in writing, that they will never be criticised for delaying a payment to verify it, even when the "CEO" is shouting. Then leaders need to follow the process themselves.

5. Use your bank's name check

Australian banks began rolling out Confirmation of Payee in 2025. It checks the account name you enter against the BSB and account number and tells you whether it's a match, a close match or no match. A "no match" on an urgent new payee is a big red flag.

6. Lock down the email accounts behind it

Many deepfake calls are paired with an email from a compromised mailbox to make the story hang together. Phishing-resistant MFA, alerts for suspicious inbox rules and managed detection and response cut off that half of the attack. Our Microsoft 365 security assessment checks the settings.

Then train for it. User awareness training should now include voice and video impersonation, not just phishing emails. Law firms and conveyancers are prime targets; we've written about that in the conveyancing scam hitting law firms, and the same tactics hit accounting practices too.

If money has already gone

Speed matters more than anything. Call your bank immediately and ask them to try to recall the payment; the first hour gives the best chance. Then report it through ReportCyber at cyber.gov.au, preserve the call logs, messages and emails, and contact IDCARE for support. If customer data was shared, you may also need to assess the incident under the Notifiable Data Breaches scheme.

Frequently asked questions

What is a deepfake CEO scam?

It's a fraud where criminals use AI to clone a senior executive's voice, and sometimes their face on video, to instruct staff to make urgent payments or share confidential information. It's an evolution of business email compromise.

How much audio do scammers need to clone a voice?

Not much. Modern tools can produce a convincing imitation from short clips, and executives' voices are often freely available in videos, podcasts, webinars and voicemail greetings. Assume any public figure in your business can be cloned.

How do you detect a voice cloning scam?

Listen for robotic rhythm, unnatural pauses and odd background noise, but don't rely on it. The reliable test is process: hang up and call the person back on a number you already have, or ask a question only they could answer.

How can businesses prevent deepfake payment fraud?

Require callback verification on known numbers for all payment and bank-detail changes, two-person approval for payments and new payees, Confirmation of Payee checks, and a clear policy that staff will never be penalised for verifying a request.

Local IT and cyber security support across NSW

Chewing IT runs managed IT and cyber security for small and mid-sized businesses from our Wyong and Hornsby offices, covering the Central Coast, Newcastle, Lake Macquarie and Hornsby.

Would your finance team catch a cloned voice? Talk to us about payment fraud protection — training, verification processes and email security that stop the transfer.

Want this handled for you?

Chewing IT looks after IT, cloud and cyber security for businesses across the Central Coast, Newcastle and Sydney. Free, no-pressure consultation — no contracts, no jargon.