BLOG

For years the standard advice on payment fraud was simple: if an instruction arrives by email, pick up the phone and check. That advice is now incomplete. Voice-clone and deepfake fraud means the voice on the phone, and even the face on a video call, can be fabricated convincingly enough to fool someone who knows the real person well. For family offices, executives and wealthy families, where a few senior people can move large sums quickly, seniority itself has become the attack surface.

This guide explains how these attacks typically unfold, the red flags that appear even when the voice sounds right, and a verification protocol a family or private office can adopt this week. It ends with what to do in the first hour if a payment has already gone out.

Two cases that changed the conversation

In September 2026, Reuters reported, citing two sources, that fraudsters stole about €95 million from Fideuram, the private banking arm of Intesa Sanpaolo. According to the report, the scheme began in February with a WhatsApp message that appeared to come from the group’s chief executive, asking for urgent help with an overseas transaction. A follow-up phone call appeared to come from a senior partner at a prominent law firm confirming the instruction, and the sources said AI was used to replicate the lawyer’s voice. Transfers went to foreign accounts, mainly in China and Hong Kong. Roughly €53 million was later recovered through cooperation between authorities in several countries; about €36 million remained missing after being moved through overseas accounts and converted into cryptocurrency. The bank declined to comment.

The lesson is not that the chairman was careless. He did what most people are taught to do: he sought confirmation. The fraud simply supplied the confirmation, in the voice of someone he trusted.

An earlier case shows the same pattern on video. In 2024, CNN reported that engineering firm Arup confirmed it was the victim of a scam in Hong Kong in which an employee paid out about US$25 million. Hong Kong police said the employee had initially suspected a phishing message about a secret transaction, but set those doubts aside after a video conference in which, as it turned out, every other participant was a deepfake recreation of colleagues.

Governments have issued similar warnings. In May 2025 the FBI warned of an ongoing campaign using text messages and AI-generated voice messages to impersonate senior US officials, and advised people to independently find a number for the person and call back to verify.

How voice-clone and deepfake fraud works

Most attacks follow a recognisable sequence:

• Research. Attackers study who holds authority, who executes payments, which advisers are trusted and how the principal speaks. Interviews, conference recordings, podcasts and social media provide voice and video samples.
• Pretext. A first message, often on WhatsApp, a new number or a personal email, introduces a confidential and urgent matter: an acquisition, a legal settlement, a regulatory issue, a family emergency.
• Corroboration. A second channel “confirms” the story: a cloned voice call from a lawyer or banker, a short video call, forged documents or letterheads.
• Pressure. Secrecy is requested, normal approvers are bypassed “just this once”, and deadlines are tight.
• Extraction. Funds are split across several transfers to new beneficiaries abroad and moved on quickly, sometimes into crypto-assets.

Red flags, even when the voice sounds right

• A request to keep the matter from people who would normally be involved.
• A new number, a new messaging app or a personal account used for a business instruction.
• Urgency tied to a deadline that leaves no time to check.
• New beneficiary accounts, new jurisdictions, or a payment split into several transfers.
• Corroboration that arrives from the same side as the request, rather than from a channel you initiated.
• Reluctance to answer a personal question, switch channels, or wait for a call-back.
• On video: short calls, poor connection used as an excuse, little interaction, cameras that “fail” when asked to do something unscripted.

Detection by ear or eye is unreliable. The FBI itself notes that AI-generated voices can sound nearly identical to the real person. The defence has to be procedural.

A verification protocol for families and private offices

1. Call-back rule

No instruction to move money, share credentials or release sensitive documents is acted on until it has been confirmed by calling the person back on a number already held on file, never one supplied in the message. The person receiving the request initiates the call.

2. Family and office code words

Agree a private verification phrase, or a small set of challenge questions, among the principal, family members and key office staff. Keep it off email and messaging apps, change it periodically, and change it immediately if anyone may have been exposed. Answers should not be discoverable online.

3. Dual approval for payments

Any payment above an agreed threshold, and any payment to a new beneficiary or a changed account, requires two independent approvers. Neither may be the person who received the request. Seniority does not override this rule, and the principal should say so in advance, in writing, so staff are protected when they refuse.

4. Pre-agreed channels

Decide which channels may carry instructions and which may not. A rule such as “we never authorise payments by WhatsApp or voice note” removes much of the attacker’s leverage.

5. A cooling-off period

For unusual or large transfers, build in a short mandatory delay. Genuine transactions can almost always absorb it; fraud depends on not having it.

6. Brief advisers and banks

Tell your private bank, lawyers and accountants how you verify instructions and ask them to follow the same rules. Attackers often impersonate exactly these advisers.

7. Reduce the raw material

Review how much of the principal’s voice, schedule and relationships is publicly available. It will never be zero, but less exposure makes convincing impersonation harder. Our family office security advisory work often starts here.

The first hour if you have been hit

1. Call your bank immediately using the number you hold, ask for the fraud team, and request a recall of the transfer. Speed matters: in the Fideuram case, quick detection and inter-bank cooperation reportedly helped recover more than half the funds.
2. Stop further payments and freeze any pending instructions linked to the same request.
3. Preserve evidence. Keep messages, call logs, numbers, recordings, documents and beneficiary details. Do not delete the conversation.
4. Report to the police and the relevant cybercrime authority in your jurisdiction.
5. Warn the people who were impersonated and everyone who might receive the next attempt.
6. Secure accounts that may have been accessed, and change code words.
7. Control communications. Decide early who needs to know and what is said. A discreet crisis and high-risk advisory response protects both recovery efforts and reputation.

Frequently asked questions

Can a voice really be cloned from public recordings?

Yes. Authorities including the FBI warn that AI-generated audio is being used to impersonate public figures and personal contacts, and that it can be very difficult to distinguish from the real voice.

Is a video call enough to verify someone?

No. In the Arup case, police said every other participant on the video conference was fabricated. Verification should rely on a call-back you initiate and on pre-agreed procedures.

What is out-of-band verification?

Confirming an instruction through a separate channel that you control, such as calling a known number on file, rather than replying on the channel the request came through.

Who in a family should know the code word?

Only those who could plausibly request or execute sensitive actions: typically the principal, spouse, adult children and a small number of trusted office staff.

How often should the protocol be tested?

At least once a year, and after staff changes. A short, announced exercise is enough to show whether people will follow the call-back and dual-approval rules under pressure.

8Dimension works with a small number of families, principals and private offices by referral. If you would like a discreet review of how your office verifies instructions, or support with family office advisory matters, please get in touch.