Risk Guardops • Updated November 2026
📝 Security Controls • 8 min read

Eight Deepfake Payment Fraud Mistakes, and the Controls That Actually Stop Them

A cloned CFO on a video call can beat an approval chain built for email. Here are eight treasury mistakes, the control that fixes each, and the policy wording that can leave the loss uninsured.

SPONSORED CONTEXTUAL PLACEMENT
💼
Evaluating Enterprise Deployments in 2026?

Request our confidential total cost of ownership matrix and vendor RFP templates.

Launch Calculator →

Why a familiar face on a call no longer proves anything

If your finance team would release a payment because the CFO asked for it on a video call, you have exactly the gap that deepfake payment fraud is built to exploit. Voice cloning can work from a short sample of public audio, and earnings calls, conference panels and podcasts supply plenty. Real-time face-swap tools can put a convincing executive into a Teams or Zoom meeting. In the widely reported 2024 case at engineering firm Arup, an employee in Hong Kong reportedly transferred a sum in the region of US$25 million after a video meeting populated with fabricated colleagues. A call with several people on it is not a safeguard.

The fix is mostly process, not technology. Below are eight mistakes that keep showing up in treasury and accounts-payable workflows, each with the control that closes it. The last one covers where crime and cyber policies may decline to pay because your own employee authorized the transfer.

Advertisement
RECOMMENDED SOLUTION

Looking to streamline vendor RFP selection?

Access pre-negotiated tier discounts through our independent institutional network.

Compare Pricing →

Accepting a recognizable voice or face as authentication

Deepfake CEO fraud works because staff check the person when they should be checking the channel. A clerk who has spoken to the CFO a hundred times will trust that voice, and a clone that reproduces the accent, pace and habitual phrases earns that trust automatically.

What to do instead: state in your payments policy that voice, video, email and chat can never authenticate a payment instruction, a bank-detail change or an exception, however senior the requester. Authentication comes only from an out-of-band check on a channel the requester didn't choose. Have the CEO and CFO sign the policy and circulate it, so a junior employee can point to it when a "CEO" on the line gets impatient.

Calling back the number the requester supplied

Plenty of companies claim to do payment callback verification, yet they dial the number in the email signature or on the caller ID, or ring the "assistant" the fake executive named. The attacker can control every one of those numbers, and caller ID is easy to spoof.

What to do instead:

  • Take callback numbers only from the vendor master file or the HR directory, as captured at onboarding. Never use a number from the request itself.
  • For executives, call a known mobile number or use an internal channel where identity is verified through single sign-on. Better still, have someone other than the recipient of the request make the call.
  • Record who called, which number they used, where the number came from and when. You'll want that record for the insurance claim.
  • Above your threshold, require a second person to make the callback.

Letting secrecy and urgency override the process

The script rarely changes: a confidential acquisition, a regulatory matter, a deal that must close today, a payment to a new beneficiary in an unfamiliar country, and an instruction not to tell anyone. Secrecy cuts the employee off from the colleagues who would catch the problem.

What to do instead: have leadership say in advance that no executive will ever ask anyone to skip verification or hide a payment from finance. Any request for confidentiality should go straight to escalation. Give staff written permission to delay, such as a 24-hour hold on any first payment to a new beneficiary, and make clear that holding a legitimate payment will never get them in trouble. Shared code words help small teams as an extra layer, but they leak and get reused. Don't let them replace the callback.

RECOMMENDED SOLUTION

Looking to streamline vendor RFP selection?

Access pre-negotiated tier discounts through our independent institutional network.

Compare Pricing →

Single-approver releases and thresholds that are easy to split

If one person can release funds, one convincing call is all an attacker needs. Per-payment limits also invite splitting. Say a 40-person company lets a single approver release anything under US$50,000. A cloned "CFO" asks for three payments of US$45,000 in one afternoon, and none of them reaches the dual-approval limit.

What to do instead: enforce dual authorization in the banking portal or ERP; a policy document alone won't do it. The second approver must not have been on the call or the email thread. Apply thresholds to the total paid per beneficiary per day and per week, not to each payment. Keep the right to add beneficiaries separate from the right to release payments. If your finance team is only two people, make a director or your external accountant the second approver through the bank's mobile app.

Changing beneficiary bank details on the strength of a call

Fraudsters clone the voices of suppliers and customers, not only executives. A call from a familiar "account manager" announcing a new bank is a common route to redirecting genuine invoices, and the loss often isn't noticed until the real supplier chases payment weeks later.

What to do instead: put bank-detail changes through their own workflow: a written request, a callback to the contact already in the master data, a cooling-off period of a few business days before the first payment, and a small test payment that the vendor confirms through that known contact. Use payee name-check services wherever they exist. The UK has Confirmation of Payee, and the EU has been rolling out Verification of Payee for euro credit transfers. India's UPI shows the beneficiary name before payment, and Brazil's Pix does the same. Several other markets, Australia among them, have introduced or are introducing similar schemes. Availability and scope vary, so check what your bank supports. Cross-border wires usually have no such check, so international payments need the manual steps every time.

Training staff on phishing emails but never on a live call

Most awareness programs still focus on suspicious links. Finance, treasury, executive assistants and the IT help desk (which handles password resets by phone) now need practice resisting someone who sounds exactly like their boss. Assume your executives' voices are already out there; limiting public audio isn't realistic for listed companies.

What to do instead: run voice and video impersonation simulations for high-risk roles at least once a year. Rehearse a short script: "I'll call you back on your directory number," then hang up. Publicly recognize staff who stop a suspicious request. Typical 2026 market ranges, which vary by contract and region: broad awareness platforms with vishing modules cost roughly US$15 to US$50 per user per year, and a targeted deepfake simulation engagement from a specialist firm often runs US$10,000 to US$50,000. Get quotes from vendors.

Buying detection tools before fixing the workflow

Detection products such as Pindrop for voice channels and Reality Defender for media analysis are useful, but they produce probabilities, not proof, and the generators keep improving. Detection also can't see a WhatsApp call to an employee's personal phone, which is where many of these attacks happen.

What to do instead: treat detection as an extra layer. Buy it first only if you handle high volumes of inbound voice, such as a contact center, private banking desk or claims line, where callbacks can't scale. Enterprise pricing is usually quoted per deployment and varies widely with call volume and integration; expect a sales process, not a price list. For most mid-size companies, callbacks to master-data numbers plus enforced dual control prevent more loss per dollar spent.

Assuming your policy pays for a transfer your staff approved

This is where many buyers find out too late that they weren't covered. Impersonation fraud cover usually comes as a specific social engineering endorsement, and many policies exclude "voluntary parting" losses unless that endorsement is in place. Wordings differ by insurer and country, so read your own.

Coverage partWhat it typically responds toCommon gap for deepfake losses
Cyber: funds transfer fraudFraudulent instructions sent to your bank after a system compromiseOften excludes transfers an employee knowingly authorized
Crime: computer fraudUnauthorized entry into or alteration of your systemsAn authorized user making a payment may not count
Social engineering endorsementEmployee tricked into sending funds by an impersonatorSublimits; required verification steps; wording limited to "email"

Social engineering sublimits commonly start in the low six figures in USD, with higher limits available at higher premium. These are typical 2026 market ranges and vary by insurer and region. Many endorsements require a documented callback before payment, and some define the fraudulent instruction as electronic or written, which may leave out a phone or video instruction. In the UK and much of Europe this cover often sits in crime or fidelity policies. In parts of Asia, the Middle East and Latin America it can be harder to get. Ask a licensed broker to confirm how your wording applies.

Next step: pull your last 20 payments above threshold this week and check whether each had a documented callback to a master-data number. Then send your wordings to your broker with three questions: does a voice or video instruction trigger cover, what verification does the policy require, and what is the sublimit?

Frequently Asked Questions

Will our cyber insurance cover a deepfake CEO fraud loss? +
Are code words between executives and finance staff a good defense against voice cloning? +
What should we do in the first hour after a suspected deepfake payment? +
RG
Risk Guardops Editorial Team
Independent research desk
Guides on Risk Guardops are produced by our editorial team. Pricing, features and regulations change often - confirm details with the vendor or a licensed professional before making decisions.
SPONSORED 2026 Vendor Pricing Benchmarks: Model exact seat pricing, setup fees, and ROI across our verified database.
Launch Calculator →