The classic version of this fraud has been around for years — an urgent call or email, apparently from a director, asking someone in finance to make an emergency payment or share sensitive information right now, no time to check. What's changed is the evidence available to the attacker. A short clip of a CEO speaking at a conference, in a podcast interview, or in a video posted to LinkedIn is now enough raw material to generate a synthetic voice that's very difficult to distinguish from the real thing over a phone line.
This isn't a hypothetical future risk. Cases of AI-cloned voices being used to authorise fraudulent payments have already hit businesses of all sizes, and the barrier to carrying out this kind of attack has dropped from "requires specialist skill" to "a free or cheap tool and a few minutes of audio."
How the Attack Actually Works
The pattern is a modern variant of a well-established fraud technique, sometimes called "CEO fraud" or business email compromise, now with audio added:
- An attacker gathers publicly available audio of a senior person at the target business — a webinar recording, a conference talk, a company video, even voicemail greetings.
- AI voice cloning tools generate a synthetic version of that voice, capable of speaking new sentences the real person never said.
- The attacker calls or leaves a voice message for someone with financial authority — often in finance or accounts — using the cloned voice, creating urgency: an emergency supplier payment, a confidential acquisition needing funds moved quickly, anything that discourages the target from asking too many questions.
- The request typically arrives alongside other pressure — a supporting email from a spoofed or compromised account, a stated reason why the "real" person can't be reached to confirm.
Why finance and accounts teams are targeted specifically: They're the people with the authority and access to move money, and the culture in many finance functions is built around responsiveness to leadership requests — exactly the instinct this attack is designed to exploit.
Why "It Sounded Right" No Longer Means Anything
The old advice to trust your ear doesn't hold up against current voice cloning technology. Tone, accent, common phrases, even characteristic pauses can be replicated convincingly enough to fool someone who knows the real person well, especially over a phone line's compressed audio quality, and especially when the listener is already primed by urgency to not scrutinise too closely. The recognition instinct that used to be a reasonable defence is no longer one you can rely on.
The Defence Isn't Technical — It's Procedural
You can't train staff to detect a good voice clone by ear. What actually works is removing voice — and email — as a sufficient basis for authorising anything sensitive, no matter how convincing or urgent it sounds.
- A callback verification rule for any financial request: Any request to move money, change bank details, or share sensitive data gets verified via a call back to a known, pre-saved number — never a number provided in the request itself, and never by continuing on the same call.
- A second channel, always: Verify through a different communication method than the one the request arrived on. A voice call gets confirmed by text or in person; an email gets confirmed by a phone call.
- A shared code phrase for genuine emergencies: Some businesses set up a simple, private verification phrase known only to senior staff and finance, changed periodically, specifically for situations where speed genuinely matters.
- No exceptions for seniority or urgency. The entire point of this attack is to create a scenario where questioning the request feels awkward or insubordinate. A policy that treats verification as mandatory — including for messages that appear to come from the most senior person in the business — removes that pressure from the individual having to make the judgement call in the moment.
- Clear authorisation limits on payments that can't be overridden by a single urgent instruction, regardless of who it appears to come from.
What to Do If You Suspect a Call Wasn't Genuine
Don't continue the conversation trying to catch the caller out — end it, and verify independently using a known contact route. If a payment was made, contact your bank immediately; fraudulent transfers can sometimes be recalled if reported within hours rather than days. Report the incident to Action Fraud, and let your IT provider know so they can check for any related compromise, such as a hacked email account that may have supplied the attacker with context.
Questions worth asking internally: Do we have a mandatory callback-verification rule for any request to move money or change bank details? Does that rule apply without exception, regardless of who the request appears to come from? Would our finance team know exactly what to do if they received a call like this today?
The Bottom Line
Voice used to be a reasonably reliable way to confirm you were talking to the right person. It no longer is, and treating it as proof is now a genuine business risk. The fix isn't better listening — it's a verification process that doesn't depend on anyone's ability to spot a fake in the moment.