AI voice cloning: how scammers use it on Australian families
Three seconds. That’s roughly how little audio some current AI voice tools need to produce a passable clone of someone’s voice, according to vendor documentation published by several commercial voice-cloning platforms now on the market. Three seconds is less than it takes to say “hey, it’s me”. A voicemail greeting covers it. So does a birthday video posted to Facebook, a podcast appearance, or a TikTok where your mum says your name. That’s the starting point for understanding why this scam has moved from theoretical to something Tech Geek now hears about from readers almost as often as we hear about phishing texts.
We’re not writing this to frighten anyone into switching off their phone. Our take is that the panic around AI voice cloning has slightly outrun the practical fix, which is cheap, boring and doesn’t involve an app at all. We’ll get there. First, the mechanics, because understanding how the trick works is most of the defence.
How voice cloning actually works
Modern voice-cloning tools are built on the same family of machine-learning models that power AI chatbots and image generators, trained on enormous libraries of recorded speech until they learn the relationship between written text and the sound of a human voice. Point one of these tools at a short sample of a specific person’s voice, and it can extract their particular pitch, pacing, accent and vocal quirks, then generate new sentences in that voice that the person never actually said.
The output isn’t flawless. Early versions a few years back had a flat, slightly robotic quality that was fairly easy to pick if you were listening for it. The current generation is better, particularly over a phone line, where background compression and a dodgy connection hide a lot of sins. A scammer doesn’t need a clone good enough to fool a sound engineer. They need one good enough to fool a frightened parent for the twenty seconds it takes to panic and transfer money.
The technology has crossed from “possible” to “trivially available”, and that’s the only fact that really matters for how we should all behave now.

The family emergency scam in practice
The scam itself is old. It’s the “grandparent scam” or the “hi mum” scam, running in various forms for years before AI entered the picture, usually via a text message pretending to be a child with a broken phone, asking for urgent funds. What’s changed is that the scammer can now attach a voice to the message.
A typical run goes something like this. The target gets a call, often late at night or during a stressful window like school pick-up, and hears what sounds exactly like their son or daughter, crying, saying they’ve been in an accident or arrested or have lost their phone and wallet overseas. A second voice cuts in, claiming to be a police officer, lawyer or bail agent, and asks for an urgent transfer, sometimes to a bank account, increasingly to cryptocurrency or gift cards because those are harder to claw back.
The emotional design is the actual weapon here. Fear and urgency switch off the part of the brain that would otherwise stop and check. That’s not new to AI scams specifically, it’s the same mechanism con artists have used since long before computers existed, just with a sharper tool attached. We’d put it alongside the way early phone phishing exploited people’s trust in caller ID before anyone understood numbers could be spoofed: a new technology slotting neatly into an old trick.
Why families are the target
Older Australians are disproportionately targeted, not because they’re less tech-literate as a group, but because family emergency scams rely on a relationship where sending money without question is plausible. A parent who would happily transfer a few thousand dollars to help a child out of a genuine jam is exactly the person this scam is built for. The ACCC’s Scamwatch service has flagged family and friend impersonation scams as a recurring category in its public reporting, and the rise of AI-assisted fraud has prompted specific guidance from the Australian Cyber Security Centre on voice and video impersonation as a growing threat.
There’s a reasonable question buried in here about whether we should be more worried about younger Australians, given how much more voice and video content teenagers and young adults have floating around publicly. We reckon the honest answer is both groups are exposed, just via different mechanisms: teens through social platforms and gaming voice chat, parents through the emotional vulnerability angle. Treating this as purely a “protect grandma” issue undersells how much raw audio exists of everyone now.
Why we don’t think detection apps are the answer
Here’s the mildly unpopular view we’ll actually defend: we’re sceptical that AI-detection apps marketed to consumers, the ones promising to analyse a call in real time and flag a cloned voice, are worth building a safety plan around. Several are already on the market or in development, and in principle the idea makes sense, since cloned audio does carry statistical fingerprints a model can sometimes catch.
In practice, this becomes an arms race where the detection model is always a step behind whatever cloning model shipped last month. A false sense of security is arguably worse than no tool at all, because it teaches people to trust a green tick instead of their own judgement. The fix that actually works costs nothing and doesn’t need updating: agree on a family codeword, a specific word or short phrase unrelated to anyone’s obvious life details, and make the rule simple. No codeword, no money, no exceptions, regardless of how distressed the voice sounds.
It’s the same logic as a bank’s two-factor authentication: a second, independent channel a scammer can’t fake just by cloning one signal. We like this fix precisely because it doesn’t rely on technology keeping pace with itself. If you want the longer version of that thinking, it overlaps with a lot of what we’ve written about AI deepfakes and Australian law, where the legal system is also playing catch-up with content that looks and sounds real but isn’t.
What to do if you get the call
If a call like this comes in, the advice from Scamwatch, and from our own look at reader-reported cases, boils down to three steps.
- Hang up and call the person back on the number already saved in your phone, not a number given to you during the call.
- Never transfer money, buy gift cards or send cryptocurrency based on a phone call alone, no matter how urgent it sounds.
- Report it to Scamwatch and, if money has already moved, contact your bank immediately and lodge a report through ReportCyber at the Australian Cyber Security Centre.
None of that requires technical skill. It requires a household rule that gets followed under pressure, which is harder to build than it sounds, because pressure is the entire point of the scam.
Protecting your own voice online
There’s a limit to how much anyone can lock down their voice in 2026. If you’ve ever appeared in a wedding video, a school assembly recording, a local news grab or a podcast, there’s audio of you somewhere that a determined scammer could pull from. We’re not going to pretend the fix is “stop posting videos”, because that ship sailed for most families years ago.
What’s more useful is tightening who can actually see that content in the first place. On Android handsets, there’s a specific settings path worth checking, and we’ve laid out the full version in our guide to Android privacy settings every Australian should check, much of which overlaps directly with limiting who can scrape your public audio and video. The same logic applies to choosing a browser’s default sharing settings, covered in our comparison of Chrome vs Safari vs Firefox vs Edge privacy settings.
Worth saying plainly: none of this stops a determined scammer targeting a specific household, because public figures, local councillors and small business owners with a public voicemail greeting are exposed no matter what their privacy settings look like. The settings help against opportunistic, broad-net scraping. They’re not a wall.
The regulatory gap nobody’s closed yet
Australia doesn’t currently have a standalone law that specifically criminalises voice cloning as a category, which surprises a lot of readers who assume there must be one by now. Existing fraud, impersonation and telecommunications offences cover a lot of the actual scam conduct, and the eSafety Commissioner’s remit over online harms extends to AI-generated content used to harass or deceive, but the legal framework is still built largely around older assumptions about what counts as impersonation.
We’ve covered the wider mess of AI-generated fakery and the law in more depth, including where video deepfakes sit, in our piece on AI deepfakes and Australian law, and the short version is that legislation tends to catch up roughly two scams late. That’s not a criticism unique to Australia, it’s the standard pattern whenever a new technology outruns the statute book, the same lag we saw with early internet fraud and, before that, with cheque forgery.
On the telco side, the switch-off of 3G networks across Telstra, Optus and Vodafone/TPG has pushed scam calls increasingly toward VoIP and spoofed caller ID over standard mobile networks, something ACMA has continued to monitor as part of its broader work reducing scam calls and messages. None of that is a reason for complacency. It’s a reminder that the scam adapts to whichever channel is cheapest and least traceable at the time.
A small business angle worth knowing
It’s not only families getting targeted. We’ve heard of variations where a scammer clones the voice of a business owner or manager, then calls a bookkeeper or junior staff member asking for an urgent invoice to be paid or a bank detail to be changed. It’s the same trick wearing a different uniform, and it sits alongside the kind of fraud we’ve covered in our guide to cyber security basics every Australian sole trader needs. A simple callback policy for any payment request, verified through a second channel before money moves, defeats this version just as well as a family codeword defeats the personal one.
Our verdict on the voice-cloning threat
We don’t think every Australian family needs to live in a state of alert about this, and we’re wary of the breathless framing some coverage gives it, as though every phone call is now suspect. But the technology genuinely has changed the calculus on one specific scam that’s been running for decades, and the fix isn’t a subscription or an app. It’s a five-minute conversation at the next family dinner about a codeword nobody puts on social media. Cheap, effective, and it doesn’t need a software update when the next model ships.
— Dale Whitfield, Founding Editor, Tech Geek



