The caller sounded exactly like the principal. Same cadence, same impatience, same way of saying the assistant’s name. The request was urgent, slightly irregular, and plausible: wire the deposit today, the deal closes in the morning, I am boarding and cannot talk.
Nothing about that call was unusual except that the principal never made it.
What happened at Arup
In early 2024, an employee in the Hong Kong office of Arup, the global engineering firm, received what appeared to be a message from the company’s UK-based chief financial officer about a confidential transaction. The employee was suspicious at first. Then came a video call. The CFO was on it. So were several colleagues the employee recognized.
Every one of them was a deepfake. According to reporting by CNN and CFO Dive, the employee made fifteen transfers totaling about $25.6 million before the fraud was discovered.
The attack did not beat Arup’s controls. It beat the human instinct that says a face and a voice you know are proof of identity.
Recognition is not verification
For most of human history, recognizing someone’s voice was a reasonable way to confirm who you were speaking to. That assumption is now a vulnerability.
Voice cloning tools need a short sample of clean audio to produce a convincing clone. For a public figure, that sample is everywhere: podcasts, earnings calls, interviews, Instagram stories, a keynote on YouTube. For a family member, it is a voicemail greeting or a video posted by a cousin. Executives, athletes and creators are the easiest people on earth to clone because their voices are the product.
Video is harder but no longer exotic. Real-time face swapping on a call is commercially available and good enough to fool someone already expecting to see that person.
So the rule is simple. Who someone sounds like tells you nothing. What matters is whether the request was verified through a channel the attacker does not control.
The callback rule
Every request to move money, change payment details, share credentials or send sensitive documents gets verified out of band. That means you hang up, or close the chat, and reach the person yourself using a number or channel you already had before the request arrived.
Not the number in the signature. Not the number that just called. A number from your own contacts, saved months ago.
If the person is genuinely unreachable, the money waits. A real deal survives a two-hour delay. A fraud does not, which is why fraudsters manufacture urgency.
The rule has to be absolute. The moment there is an exception for “when it is obviously him,” the exception is what gets exploited. The Arup fraud was built specifically to clear that bar.
Code words
For families and small teams, a shared code word adds a second layer. It is a word or phrase agreed in person, never written in email or text, that the principal says when making an unusual request by phone, and that family members use when calling about an emergency.
It works because a clone can reproduce a voice but cannot know a secret that was never recorded. Keep it boring and not guessable from your public life. For executive teams, a challenge question drawn from an offline conversation does the same job: “What did we argue about at dinner Tuesday?” beats any password.
Briefing the people who actually move the money
Principals rarely send the wire. Assistants, bookkeepers, business managers and spouses do, and they are the ones who feel they cannot say no to the boss. The attack depends on that.
The briefing should give them explicit permission, in writing, from the principal:
- You will never be penalized for delaying a transfer to verify it, even if it turns out to be real and even if I am annoyed in the moment.
- Any request for payment, new banking details, gift cards, crypto or credentials is verified by callback to a saved number. No exceptions, including me.
- Urgency, secrecy and “do not loop in anyone else” are the three signals of fraud. Any one of them triggers a callback. All three together triggers a call to me and to security.
- A video call is not verification. Treat it the same as a phone call.
Run a drill once. Have someone call the assistant with a plausible pretext and see what happens. The goal is to make the callback reflex real before it is needed.
For family, the conversation is shorter. Grandparents and teenagers are both targets of “I am in trouble, send money, do not tell anyone” calls. Teach them the code word and the rule that any emergency request gets a call back to the person’s own phone.
The first 30 minutes if money moved
Speed matters more than anything else. Funds often sit at the receiving bank briefly before being moved on.
- Call your bank’s fraud line immediately. Ask for a wire recall. Give them the amount, the beneficiary account, the time and the reference number. Do this before anything else.
- File with the FBI’s Internet Crime Complaint Center at ic3.gov. For large domestic wires reported quickly, IC3’s Recovery Asset Team can work with banks to try to freeze funds. Include every detail of the transfer in the complaint.
- Preserve everything. Do not delete the email thread, the call log, the voicemail or the chat. Screenshot the video call participant list if it still exists. Note exact times. This becomes evidence and supports any insurance claim.
- Notify the principal and counsel. Insurance and disclosure clocks may start now.
- Assume the attacker is still inside. Deepfake fraud is often preceded by a compromised mailbox used to study how the company talks. Rotate credentials for the people involved and check for mail forwarding rules.
Where ShutOut comes in
We write the verification protocol with the principal and then install it with the people who actually hold the checkbook: assistants, business managers, family office staff, spouses. We run the drills, build the code-word plan, and audit the mailboxes attackers would study before making the call. When money has already moved, we work the recall alongside your bank and counsel and handle the evidence so it holds up later.