The Scam That Sounds Exactly Like You

Your bookkeeper takes a call at 4.40 on a Friday. It is you. Your voice, your cadence, the way you always open with “mate, quick one”. You are between meetings, the supplier’s bank details have changed, and the payment has to go today or Monday’s delivery does not land. She has heard you sound exactly like this a hundred times.

It is not you.

Three seconds of recorded audio is enough to build that voice. You have published a lot more than three seconds. Every podcast appearance, every video, every voice note, every webinar somebody recorded. I have hours of my own voice online and so does anyone who has done any marketing in the last five years. That horse bolted a while ago, and I would not want it back. Publishing is how people find you.

So the defence has to sit somewhere else.

What Australia Is Actually Reporting

Payment redirection scams took $166.8 million from Australians in 2025, up more than 9 per cent on the year before, which made them the second biggest category by reported loss behind investment fraud. That is the scam where a business is convinced to send money to the wrong account, usually through a change of bank details on an invoice that otherwise looks completely ordinary.

The Australian Signals Directorate puts email compromise at the top of the threats reported by Australian businesses in its most recent Annual Cyber Threat Report. Business email compromise with a direct financial loss accounts for around 15 per cent of business cybercrime reports, with another 19 per cent covering email compromise without a direct loss. Roughly one in three cybercrime incidents hitting an Australian business starts in an inbox.

Voice is the newer layer sitting on top of all that. Reported losses from AI voice cloning scams in Australia reached $25.8 million in the first half of 2025 alone, and Deloitte has projected global losses from AI-enabled fraud climbing towards US$40 billion by 2027.

Every one of those figures is understated. Businesses that get taken this way frequently never report it. It is embarrassing, they assume nothing will come of it, and the loss lands on a small company that just wants to get through the week.

Why A Small Business Is The Better Target

Coverage of this tends to focus on the spectacular cases, the multinational that wired tens of millions after a video call with a fake finance director. I think SMEs are the easier mark, for three fairly boring reasons.

There is usually no written verification procedure for payment requests. There is often one person who can move money on their own. And there is a culture where a direct, urgent instruction from the owner gets carried out, because that is exactly what makes a small business quick.

A large company has dual authorisation, a payments team and a fraud policy somebody is paid to maintain. Your business has Rachel, who has been with you nine years, who knows your voice better than your partner does, who would never dream of questioning you, and who can move $80,000 before lunch.

Speed is the whole advantage of a small business. It is also the attack surface.

Detection Is Losing. Process Is Holding.

Owners usually ask me which tool spots a cloned voice. The honest answer is that cloning is improving faster than detecting. Some tools help at the margins. None of them are a control I would want to bet a quarter’s profit on.

What works is older and duller than any product. A rule, agreed in advance, applied every time, with no exception for the boss.

• Call back on a number you already hold. Never the number that rang you, never the one in the email signature. The number in your own contacts, or on the signed contract. A cloned voice does not survive being hung up on and rung back.
• Agree a passphrase offline. Something ordinary that has never appeared online, used whenever money or credentials are in play. Change it when someone leaves.
• Two people above a threshold. Pick an amount that would genuinely hurt to lose and require a second approver above it. Fooling two people at once is a much harder job than fooling one.
• Verify every bank detail change by phone. A new account number on a familiar invoice is the highest-risk event that happens in your business. Ring the supplier on the number you have had for years and read the digits back to them.
• Treat urgency as the signal. Almost every one of these stories has a deadline in it. Real suppliers cope with a two-hour delay. Slowing down costs nothing and breaks the entire play.
• One page, then a drill. A policy nobody has read is decoration. Once a year, send your own team a realistic fake request and see what happens. Fix the gap without blaming the person who found it.

Your Team Has To Be Allowed To Say No To You

This is the part that gets skipped, and it decides whether any of the above is real.

A verification rule only holds if the person applying it is genuinely safe to apply it to you. If you have ever been short with someone for slowing you down, if you wave away a process when it lands on the owner, if the unspoken rule is that your urgent request jumps the queue, then your rule is decorative. On the Friday it matters, that person will weigh two minutes of awkwardness against your mood, and they will pay the invoice.

So say it out loud, in front of everyone. If a request to move money comes from me and something feels off, ring me back on my usual number. Every time. If I sound annoyed about it, ignore that, because it probably is not me.

This is a leadership problem before it is a security problem. Businesses where every decision routes through the owner are more exposed than businesses where authority is distributed and questioning is normal, which is one more reason getting yourself out of the middle of everything matters well beyond your own workload.

If It Happens Anyway

Move in hours, not days. Ring the bank straight away and ask them to attempt a recall, because the first few hours are the window where money is realistically recoverable. Report it through ReportCyber and Scamwatch. Tell your insurer if you hold cyber or fraud cover. Then check the email account for forwarding rules or filters somebody else set up, because payment redirection usually begins with weeks of quiet reading before anyone asks for a cent.

And tell whoever made the payment that they are not in trouble. That one sentence buys you speed the next time, and speed is the only thing that gets money back. A team afraid of being blamed reports late, and late is when the money is already gone.

The Habit Underneath It

There is a thread running from this back through everything I write about AI. A cloned voice is a confident output that happens to be false. So is a model inventing a citation, or a chatbot telling your customer about a refund policy you have never offered. The skill is the same in both directions: check before acting on something purely because it sounded right.

Build that instinct into how your business uses AI and you are protected on both sides, from the AI you use and the AI used against you. Put it in your AI policy while you are there. Most policies cover what staff may paste into a chatbot and say nothing at all about what to do when the chatbot, or the voice on the phone, is the one attacking you.

What I Actually Want

I am not going to stop publishing my voice, and neither should you. The recordings are how people work out whether they want to deal with you. Pulling them down would cost far more than the fraud ever will.

What I want is a business where the person who moves the money feels completely free to hang up on me and ring back on the number they already have. Two minutes of mild awkwardness, every time, forever. It is the cheapest insurance you will ever buy. The only hard part is the day you tell your team you mean it, and then proving you meant it the first time they use it on you.

Frequently Asked Questions

What is an AI voice cloning scam?

An AI voice cloning scam uses a synthetic copy of a real person’s voice to make a fraudulent request sound legitimate. In a business setting it usually involves a cloned owner or executive calling someone who can move money, creating urgency, and asking for payment to a new account. It is often paired with a compromised or spoofed email account, so the call appears to confirm an instruction the target has already seen in writing.

How much audio does someone need to clone a voice?

Current tools can produce a convincing clone from as little as three seconds of recorded speech. Any podcast appearance, video, webinar recording, social clip or voicemail is more than enough. For most business owners who do any marketing at all, assume a usable sample of your voice is already public and build your defences on that assumption.

How do I protect my business from payment redirection and voice-clone fraud?

Use process controls rather than relying on detection. Call back on a number you already hold rather than one supplied in the message, agree a verbal passphrase offline for money and credential requests, require two people to approve payments above a set threshold, and verify every change to supplier bank details by phone using a long-standing contact number. Treat urgency as a warning sign, write the rules on one page, and test them with a realistic drill.

What should I do if my business pays a scam invoice?

Act within hours. Contact your bank immediately and ask them to attempt a recall, because the earliest hours are when recovery is most likely. Report the incident through ReportCyber and Scamwatch, and notify your insurer if you hold cyber or fraud cover. Check the affected email account for forwarding rules or filters set up by an intruder, since payment redirection often begins with weeks of quiet mailbox access. Tell the person who made the payment they are not in trouble, because fear of blame delays reporting.

Can AI detect a cloned voice?

Detection tools exist and some of them help, but cloning quality is improving faster than detection, so no tool should be the control your business depends on. A verification step agreed in advance and applied every time, including to the owner, is far more reliable than asking someone to judge whether a voice on the phone is real.