Your CFO's Face Is Now an Attack Surface
November 12, 2026 · 6 min read
The deepfake fraud conversation keeps getting framed as a detection problem. Can your people spot the fake? Can your tools?
Wrong question.
In early 2024 a finance employee in Arup's Hong Kong office joined a video call with the company's CFO and a few colleagues. He'd been suspicious of the email that set it up. The call put his doubts to rest, because everyone on it looked right and sounded right. He made 15 transfers totalling about US$25.6 million. Every person on that call was fake.
People still talk about Arup like it was a freak event. It wasn't. It was a proof of concept, and the criminal market took notes.
The numbers stopped being theoretical
In September, Gartner published a survey of 297 senior security leaders. 41% had at least one social engineering incident involving a deepfake on an employee audio call in the past year. 36% saw one on a video call.
That's not a fringe problem. That's close to half.
The FBI started tracking AI-enabled fraud in 2025 and logged $893 million in reported US losses in its first year. Deloitte expects US losses from generative AI fraud to reach $40 billion by 2027, up from $12.3 billion in 2023.
Canada's numbers look smaller on paper. They aren't. The Canadian Anti-Fraud Centre recorded nearly C$3 million lost to family-emergency and grandparent voice-cloning scams in 2024, and the CAFC and RCMP estimate reports capture only 5 to 10% of actual fraud. Do the math.
What changed is the price
Voice cloning used to need studio time and a lot of clean audio. Now a convincing clone takes about three seconds of recorded speech. Tools that do it are sold online for less than $5 a month, and many only want around 30 seconds of audio.
Think about how much of your CEO's voice is sitting on YouTube right now. Earnings calls, conference keynotes, that podcast they did last spring. Your execs have been training the attacker's model for free.
Video is on the same curve. Palo Alto's Unit 42 had a researcher with no deepfake experience and a five-year-old computer build a working synthetic identity for job interviews in 70 minutes.
Seventy minutes.
The scam itself didn't change. Fake boss, urgent wire. CEO fraud and the grandparent scam are decades old. What changed is what it costs to run a believable version a thousand times. Fraud crews are businesses, and when the cost of a convincing impersonation drops to near zero, they run more of them. Pindrop watched deepfake attempts in the contact centres it monitors go from roughly one a month to seven a day over 2024.
It's not just the wire transfer
Most coverage stops at the fake CFO. That's the headline. The quieter damage is happening in places nobody budgets for.
Start with your help desk. Phone-based password and MFA resets were built on the assumption that only you sound like you. That assumption is dead. Canada's banking regulator OSFI has flagged AI voice cloning as a fraud risk against bank employees and their customers. If your reset process is "caller sounds right and knows the employee number," you have a hole.
Then your hiring pipeline. North Korean IT workers have been landing remote jobs under fake identities for years. In July, eleven countries issued a joint advisory describing operatives using real-time video manipulation in interviews. Hire one by accident and you've handed a hostile state a laptop and a paycheque. You may also have a sanctions problem your lawyers won't enjoy.
And at home. The grandparent scam with a cloned voice is ugly and it works. A lot of executives will meet this threat for the first time at 11 p.m., with what sounds like their kid crying on the line.
A little perspective
The sky isn't falling. Gartner's own analyst points out that most attacks will still lean on stolen credentials and weak recovery processes. Deepfakes don't replace the old playbook. They make it more believable. And plenty of the scarier stats floating around come from vendor telemetry, not a census. Treat them as direction, not gospel.
Direction still matters. This line only goes one way.
Stop trying to spot the fake
Most deepfake awareness training teaches people to hunt for glitches. Odd blinking. Lip sync slightly off. Weird lighting around the hairline.
That's training for last year's models. The tells are disappearing faster than anyone can put them in a slide deck. Vendors keep selling "spot the deepfake" modules because they demo well, not because they hold up. The Arup employee was already suspicious and still lost, because the call looked fine.
So quit making your people the detection engine. Make the process the control.
- Call back on a known number. Any payment change, new payee or urgent wire gets verified by calling the person on a number you already had. Not the one in the email. Not the one they read out on the call.
- Two people for money. Over a set threshold, no single employee moves funds, no matter who's asking. Including the CEO. Especially the CEO.
- Retire the voice as an authenticator. Help desk resets need something a clone can't fake, like a push to an enrolled device or a manager approval in a separate system.
- Agree on code words. Low-tech, and it works. Set one with your exec team and another with your family. If the "CEO" on the call can't produce it, the call is over.
- Verify humans in hiring. At least one live ID check for remote hires in sensitive roles. Watch for equipment shipped to addresses that don't match the candidate.
- Give people permission to say no. This is the big one. Deepfake fraud works by borrowing authority. If your culture punishes the AP clerk who holds up the CEO's urgent wire to verify it, you've already lost. Say it out loud, from the top: verifying is never insubordination.
Bottom line
For as long as we've had phones and video calls, seeing and hearing someone was proof enough. That's done. A face is data now. So is a voice. Both can be forged on a laptop for less than a streaming subscription.
Your controls need to assume that. Most still don't.
The fix isn't exotic. It's boring process, enforced every single time, with leadership going first.
Boring wins this one.
Next Step
Ready to strengthen your organization's resilience?
A 30-minute discovery call to discuss your cybersecurity posture, incident readiness, and whether advisory support is the right fit.