Cloned voices and payment fraud: the call-back rule that stops it

The call comes late on a Friday. It's the managing partner, and he sounds rushed. A settlement needs to go out today, the client is waiting, and he's about to board a flight. He gives the account details and thanks you for sorting it out.

It was his voice. It wasn't him.

How it works

Copying a voice no longer takes a studio. A short sample is enough, and for most professionals a sample is easy to find: a webinar, a conference talk, a podcast, a recorded hearing, a voicemail greeting.

The FBI's public warning on generative AI describes criminals cloning voices to impersonate people and get access to money, and using fake video in live calls to pose as executives (FBI public service announcement). Its 2025 crime report names voice clones among the tools behind nearly $893 million in AI-related losses (FBI, April 2026).

The voice is only the last step. Before the call, the attacker has usually done the reading:

• Who handles payments at your firm, from your website or LinkedIn.
• Which matters or deals are closing, from public filings or a compromised mailbox.
• When the person they are impersonating is traveling or in court, and can't be reached.

Why professional firms are targets

Firms that hold or move client money are the obvious ones.

• Law firms hold client funds in trust and release them on a partner's instruction.
• Real estate and title firms move large sums on a fixed day, arranged almost entirely by email and phone. We cover that case in wire fraud at closing.
• Financial advisers act on client instructions to transfer money.
• Accounting practices run payroll and pay invoices for clients.

In each case, a trusted voice giving a plausible instruction is how the work normally gets done. That's what the fraud exploits.

Why software alone won't stop it

Email filtering can catch a forged message. Nothing sitting on your phone line can tell you a voice is synthetic, and you shouldn't expect staff to hear the difference. The copies are good and getting better.

The defense is a procedure. It works because it doesn't depend on detecting the fake.

The call-back rule

Any instruction to send money somewhere new, or to change where money goes, is confirmed by calling back on a number you already hold.

Four details make it work:

1. Use a number from your own records. Never one given in the email or on the call. If the request says "call me on this number, my phone's broken," that's the warning sign.
2. No exceptions for seniority. The rule protects partners, because it's their voices being copied. It needs to be the partners who announce it.
3. No exceptions for urgency. Urgency is the attacker's main tool. A real client can wait ten minutes for a call back. Write that sentence into the policy.
4. Two people for large amounts. One person takes the instruction, a second approves it.

For extra protection, agree a verification word with the people who can authorize payments. The FBI recommends the same idea for families.

Tell your clients

Half the risk sits with your clients, who may get a call or an email that appears to come from you. Put a line in your engagement letter and your email footer:

We will never change our bank details by email or phone alone. If you receive a request to pay a different account, call us on the number you already have before sending anything.

It costs nothing and it gives your client permission to be suspicious.

Put it in writing

A rule that lives in people's heads fails the first time someone is busy. Write it as a one-page policy, have everyone who touches payments sign it, and go over it in your security awareness training. A written procedure is also what a cyber insurer wants to see, which we cover in what cyber insurers ask.

We check how firms handle payment changes as part of our free IT and cybersecurity review. If you'd like the policy written for you along with the rest of your security paperwork, that's part of compliance support.