The Deepfake Phone Call From the Boss: Stopping AI Voice Fraud With One Simple Rule
Cloned voices and faked video calls are now good enough to fool colleagues. Detection tools will not save a busy finance team. A simple rule about how payment requests are checked will.
Guide. Written by Dragonfly IT. Drafted with AI help and checked against the sources listed on this page.
A phone call from a director asking for an urgent payment used to be a reasonable thing to act on. The voice was familiar, the request made sense and there was no reason to doubt it. AI voice cloning has changed that.
The good news is that the defence does not depend on spotting a fake. It depends on a rule that every payment request is checked the same way, however convincing the caller sounds.
How easy is it to clone a voice?
ICAEW, the accountancy body, noted in January 2025 that some voice cloning platforms need only around a minute of audio to produce a convincing result, and that some are free. A director's voice can often be found in a podcast, a conference video or a social media clip.
ICAEW also made a point that matters for anyone using voice as a security check: recognising a familiar voice may no longer prove who is speaking.
Is this really happening to businesses?
The best known case involved Arup, the British engineering and design firm. In 2024 a finance worker in its Hong Kong office joined a video call with people who appeared to be the company's chief financial officer and other colleagues. They were all fakes. Reports based on Hong Kong police accounts say about 25 million US dollars was sent across 15 transfers before the fraud came to light.
Insurance broker Gallagher's 2026 executive protection research, reported in April 2026, found that half of the organisations surveyed had faced at least one executive impersonation or deception attempt in the previous year, and that 51% of senior leaders named AI enabled deception as their top concern. This is broker research rather than an official statistic, but it matches the direction of other reports.
Why detection is the wrong place to start
Tools that claim to spot deepfakes exist, but a busy person taking a call on a mobile phone is not going to run audio through one. The fake only has to work once, and it will usually arrive with pressure attached: a confidential deal, an angry supplier or a deadline today.
A process does not get tired or flustered. If every payment request is verified the same way, it does not matter how good the fake is.
The call back rule
The rule is simple to state. Any request to make an urgent payment, pay a new supplier or change bank details is confirmed through a second channel before money moves. In practice that means hanging up and calling the person back on a number already held by the business, or confirming through a separate secure channel, never using contact details given in the request itself.
ICAEW's advice adds two supporting habits: confirm high value and unusual payments through more than one channel, and split duties so that more than one person is involved in setting up, approving and releasing a payment. It also suggests making it normal to call the boss back to check, so nobody feels they are being disrespectful.
What to do this week
Turn the rule into something staff can follow:
- Write the call back rule down in one paragraph and send it to everyone who can pay money or change supplier details.
- Make sure the directors say, in writing, that they expect to be called back and will never be annoyed by it.
- Keep a short list of verified phone numbers for key suppliers and senior staff, held outside email.
- Require two people for any payment above an agreed amount, and for any change to bank details.
- Stop using voice recognition alone as an identity check for password resets or account changes.
Where Dragonfly IT can help
Deepfake fraud often starts with an email that sets up the call, and ends with a payment process that had no second check. Both ends can be strengthened.
Dragonfly IT's cyber security service covers security awareness training and a review of the finance processes around payments, and the email security service covers the Microsoft 365 settings that make impersonation emails harder to send.
Frequently asked questions
How much audio does it take to clone a voice?
ICAEW reported in January 2025 that some voice cloning platforms need only around a minute of audio to produce a convincing result, and that some of these tools are free. Recordings from podcasts, webinars or social media videos are often enough, which is why senior staff are common targets.
What is a call back rule?
It is a rule that any request to make an urgent payment, pay a new supplier or change bank details is confirmed before money moves. The person hangs up and calls back on a number the business already holds, never one given in the request. It works however convincing the original caller sounded.
Can a deepfake really fool someone on a video call?
Yes. In the widely reported Arup case in 2024, a finance worker in Hong Kong joined a video call with people who appeared to be the chief financial officer and colleagues. Reports based on police accounts say about 25 million US dollars was transferred before the fraud was discovered.
Is a familiar voice still a safe security check?
Voice alone should not be relied on. ICAEW notes that voice verification may no longer be reliable now that voices can be cloned. For password resets, account changes and payments, use a second channel or a method that cannot be copied, such as an authenticator app or a known phone number.
Sources
- ICAEW: How to guard against voice cloning and deepfake scams (15 January 2025) (opens in a new tab)
- Fortune: Arup revealed as victim of a 25 million dollar deepfake scam (17 May 2024) (opens in a new tab)
- Insurance Business UK: Half of UK firms hit by executive impersonation, Gallagher research (16 April 2026) (opens in a new tab)
- Gallagher: Executive Protection Report 2026 (opens in a new tab)
