Is That Really Your Boss on the Phone? How to Stop Deepfake Scams

How deepfake scams work

Deepfake scams are a high-tech version of the phishing scams we covered in week 1. Instead of a fake email, criminals use AI to create fake audio or video of someone you trust. The goal is the same: get someone to move money, share a password or hand over data. These are the most common forms:

The source audio is easy to find. In McAfee’s Beware the Artificial Impostor research, as little as three seconds of audio was enough for a free online tool to produce a voice clone with an 85% match to the real speaker. A LinkedIn video, a podcast, a webinar or a voicemail greeting provides far more than that.

Most deepfake voice scams follow the same four steps:

1. 1Collect
   Criminals pull public audio or video of a leader or colleague from LinkedIn, YouTube, Instagram, podcasts or webinars.
2. 2Clone
   Cheap, consumer-grade AI tools copy the person’s pitch, accent and speech patterns.
3. 3Call
   Internet calling software can spoof the caller ID, so the call can even appear to come from a familiar number. The “caller” then invents a crisis.
4. 4Cash out
   The cloned voice pushes for an urgent, hard-to-reverse payment, a password reset or a way around normal approvals.

Common deepfake scam scenarios

Why deepfake scams matter for Australian businesses

Deepfake scams are no longer a future threat. They’re already being used to steal money from businesses, and the losses can be huge.

• A US$25 million lesson: In early 2024, an employee at the Hong Kong office of engineering firm Arup transferred about US$25 million after a video call in which the “chief financial officer” and several “colleagues” were all deepfakes.
• Scamwatch is warning Australians: Scamwatch warns that criminals now use AI to clone voices, create fake videos and send convincing, tailored messages.
• It’s a national priority: The Australian Signals Directorate has listed preparing for AI-enabled cyber threats as one of its priority actions for businesses this Cyber Security Action Month.
• Losses are already being counted: The FBI’s Internet Crime Complaint Center recorded 22,364 complaints with an AI-related element in 2025, with US$893 million in reported losses. That includes more than US$30 million in business email compromise scams with a confirmed AI component, often backed up by a cloned voice of a CEO or CFO.
• Most Australians can’t tell: In McAfee’s research, more than 60% of Australian respondents said they couldn’t tell a real voice from an AI-generated one on the phone, and 7% said they’d already been the victim of an AI voice scam.
• It supercharges an existing problem: Business email compromise was already one of the most reported cybercrimes against Australian businesses. Deepfake scams make those requests far more convincing.

Five warning signs of deepfake scams

You may not be able to spot a fake voice, but deepfake scams still follow familiar patterns. Pause when you notice any of these:

• Sign 01
  An unknown number or new account
• Sign 02
  Extreme urgency
• Sign 03
  Secrecy or “keep this quiet”
• Sign 04
  Money, bank details or passwords
• Sign 05
  Resistance to a call back

• An unknown number or new account: “My phone’s dead, I’m on a mate’s” or a video call from an account you haven’t seen before is a common opening line. Caller ID can also be spoofed, so even a familiar number isn’t proof of who’s calling.
• Extreme urgency: “Before 5pm or we lose the deal.” Time pressure is designed to stop you thinking and checking.
• Secrecy: Requests to keep something confidential “until it’s signed” are there to stop you talking to the real person or a colleague.
• Money, bank details or passwords: Any call that ends in a payment, a bank detail change or a request for login details deserves a second check.
• Resistance to a call back: If the caller can’t take a call back, can’t give the code word or pushes you to skip normal approvals, stop.

Notice that none of these signs depend on how the voice sounds. That’s the point. Deepfake scams are caught by the request, not the voice.

The 3-step process that stops deepfake scams

A few years ago, fake voices sounded robotic. Today, the best voice clones are very hard to tell apart from the real thing, especially over a phone line, in a noisy car or when you're under pressure. Telling staff to "listen for something odd" sets them up to fail. The better approach is to assume you won't be able to tell, and build a process that works regardless.

• Step 01
  Hang up and call back on a known number
• Step 02
  Ask for the agreed code word
• Step 03
  Get two people to approve payments

• Hang up and call back on a known number: If anyone asks for money, passwords or data by phone or video, end the call and ring them on the number you already have. A real boss won't mind, and it moves the check outside the call a criminal controls.
• Ask for the agreed code word: Choose a word or phrase for urgent requests that's never written in email or posted online. If the caller can't give it, the request stops. Change it if you think it's been shared.
• Get two people to approve payments: No single person should be able to send money or change bank details on a phone request alone. Two sign-offs give everyone permission to slow down.

It also helps to tighten privacy settings on personal social media accounts, so it's harder for criminals to collect audio of staff and their families. You don't need to stop posting business videos. Just assume any public voice can be copied, and let the process do the work.

Write these steps down, share them with your team and, most importantly, make it clear that following the process will never get anyone in trouble, even if the request really was from the boss.

The other AI risk: client data in free AI tools

Deepfake scams aren't the only way AI creates risk. A quieter problem is staff pasting client details, contracts or financial information into free AI chatbots to save time. Depending on the tool and its settings, that information may be stored outside your control or used to train future AI models.

If that information includes personal details, it could become a privacy issue. Under the Privacy Act, businesses covered by the Notifiable Data Breaches scheme must notify the OAIC and affected people if personal information is exposed in a way likely to cause serious harm.

Banning AI rarely works, because people just use it on their phones. A better approach is to:

• Give staff an approved tool: Microsoft 365 Copilot works inside your existing Microsoft 365 security and permissions, rather than sending data to a consumer chatbot.
• Write a short AI policy: Spell out what can and can't be shared, which tools are approved and who to ask when unsure.
• Check permissions first: Review your Microsoft 365 sharing settings before rolling out Copilot, so it can't surface files people shouldn't see.

What to do if you're targeted by a deepfake scam

Speed matters. Acting quickly gives you the best chance of limiting the damage, though recovering money isn't guaranteed.

• Step 01
  Stop and verify
• Step 02
  Call your bank if money was sent
• Step 03
  Tell your IT provider
• Step 04
  Warn your team
• Step 05
  Report it

• Stop and verify: Don't act on the request. Contact the real person on a number you already have.
• Call your bank: If money was sent, call your bank straight away and ask whether the payment can be recalled or frozen.
• Tell your IT provider: They can check whether any accounts were compromised and help secure email, phones and payment systems.
• Warn your team: Criminals often target several people in the same business, so let everyone know what to look out for.
• Report it: Use ReportCyber or call the Australian Cyber Security Hotline on 1300 CYBER1 (1300 292 371), and report it to Scamwatch.

We'll cover incident reporting in more detail in week 4 of this series.