The CEO’s Voice Is Not a Payment Control

Print Friendly, PDF & Email
CCG Catalyst Commentary

The CEO’s Voice Is Not a Payment Control

November 18, 2026

Start with two banks in the same federal indictment, hit by the same con, with different endings. Prosecutors allege that three men called employees at companies around the world pretending to be each company’s own chief executive, with AI supplying the CEO’s voice or image. The fake CEO told the employee a confidential deal was underway, handed them off to fake lawyers at real-sounding law firms, and directed them to wire the money. Between 2024 and 2026, prosecutors say, the scheme took in roughly $129 million. Buried in the charges are two banks that got the same treatment with different endings. At one New York-area bank, an employee believed the impersonation and authorized a fraudulent wire of approximately $2.85 million – but the bank’s back office caught it, and the money never left. At another bank in Rhode Island, approximately $2.68 million allegedly left and did not come back. Same scheme, same script, two outcomes. The difference was not a smarter employee - both employees were fooled. The difference was what stood between a fooled employee and a released payment. That difference is the entire subject of this article.

What the Attack Actually Exploits

The public record does not say whether AI was used against either bank specifically, and it does not identify the control that stopped the New York wire, the defendants are accused, not convicted, and the banks are unnamed. What the indictment does describe is the mechanics, and they are older than the technology: a caller with the CEO’s voice invokes a confidential acquisition, hands the employee to a credible outside “attorney,” imposes a deadline, and this is the detail worth underlining. It used a script designed to keep the employee from discussing the call with colleagues. Executive authority, professional credibility, urgency, and secrecy. The AI strengthens the impersonation, it is not the attack. The attack is the exception, the moment an institution’s process bends because the person asking appears powerful and the matter appears too confidential to check. None of this arrived without warning: FinCEN alerted institutions to deepfake-enabled fraud in November 2024, and the FBI has told businesses for a decade to verify payment instructions through independently obtained contacts, never through numbers supplied in the request itself.

As I stated, the pattern is not new, only the production values are. In early 2024, an employee of the engineering firm Arup joined a video conference with what appeared to be the company’s CFO and several colleagues every one of them a deepfake and sent approximately $25 million across fifteen transfers before anyone verified a thing. A few months later, a caller with a synthetic version of Ferrari CEO Benedetto Vigna’s voice pressed a Ferrari executive about a confidential acquisition and the executive asked one question only the real Vigna could answer, about a book he had recently recommended. The caller hung up. One company lost eight figures to a fake it could see on screen; the other was saved by a question no deepfake could answer. Set those two outcomes beside the indictment’s two banks and the lesson repeats itself: the defense that works is not better eyes or ears. It is verification that does not care how convincing the impersonation is.

A Familiar Voice Is Not Payment Authority

Here is where community banks should resist the obvious conclusion. Your franchise is built on knowing your customers and your customers knowing you and in fraud defense that familiarity is a genuine asset: community bankers notice when something is off about a customer in ways no model does. But familiarity is a detection layer, never an authorization layer. A voice - recognized, familiar, warm - is an identity claim, and identity is only the first of four separate things a payment requires: identity, entitlement to request, approval bound to the actual details of the transaction, and release. The stopped wire in the indictment is proof that these are different events: fooling the employee and releasing the payment were separated by a control, and the control won. The customer-facing version of this discipline fits in one sentence, and I would put it in writing for every treasury client: we verify unusual payment instructions even when we know you - that is how we protect the relationship.

Big Banks Are Not the Model

The money-center banks solve this with scale with around-the-clock fraud operations, deep approval benches, release engines with segregation built in. A community bank cannot copy that staffing, and it does not need to, because the control that matters is process, not headcount. For the bank’s own payments: independent verification of unusual executive requests through contacts the bank maintains, not contacts the request supplies; separation of initiation from release; approval bound to amount and beneficiary, not to a general instruction; and every exception logged and reviewed. For business customers’ payments – the track community banks too often treat as the customer’s problem – a documented setup review of administrators, entitlements, thresholds, payee-change verification, and escalation contacts, tested jointly where the customer agrees. What the big banks’ scale actually buys is capacity, and capacity is the honest constraint to plan for: a Tennessee community banker told Congress what fraud response really costs a small institution – recovery delays, staff pulled from customers, tools like Positive Pay that help with checks and ACH do nothing for the wire leaving this afternoon. Put a capacity model next to the control model: eighty alerts at six minutes each is a full staff-day before anyone calls a customer. A control your staffing cannot operate is a policy, not a control.

Straight-Through Processing Moves the Control - Not the Duty

There is a third track, and it is the growing one: straight-through processing, where the business customer initiates, approves, and releases payments through the bank’s digital treasury tools with no banker in the loop. Read correctly, STP shifts the per-payment control decision to the end customer and with it, much of the loss allocation. Under the UCC’s security-procedures framework, when a payment order clears a commercially reasonable security procedure the customer agreed to, the loss on a fraudulent order generally sits with the customer rather than the bank, though the allocation always depends on the agreement, the procedure, and the facts. But notice what the bank traded for that protection, because the deal has two sides: the bank must offer controls that are commercially reasonable against today’s threats – dual approval with genuinely independent approvers, out-of-band verification, limits, payee-change confirmation – keep them current as the threats evolve and document the customer’s configuration choices, including the customer who declines a control, in writing. “Commercially reasonable” is measured against the fraud of 2026, not the fraud of the year the platform shipped: a toolset that was reasonable before cloned voices may not be reasonable against them, and an aging digital banking platform quietly moves the risk back onto the bank that let it age. If STP is your model – and for most community banks it should be a bigger part of it – then your product roadmap is itself a fraud control, and the vendor who maintains that platform is carrying part of your loss allocation. That also belongs in the vendor file too.

Two traps hide inside that trade-off, and both come from good intentions – two ways a bank can break its own side of the deal without noticing. The first is the random callback. A bank that offers straight-through processing but occasionally reaches in - spot-checking wires, calling back on some transfers and not others - is not adding a control; it is blurring the allocation. The moment the bank touches the wire, it owns a piece of the outcome: an undocumented, inconsistently applied review invites the argument that the bank undertook to screen the payment and did it badly, and it muddies the clean position that the agreed procedure was followed. If review is part of your procedure, write it into the agreement, define exactly when it triggers, and perform it every time it triggers. The second trap is the insurance file. Fidelity bonds and cyber-crime policies increasingly condition coverage on the verification procedures the bank described, applied consistently and a claims adjuster who finds the bank episodically touching wires it said flowed straight through has two paths to a denial: the loss arose from the bank’s own failed review, or the stated procedure was not the real one. Read the policy before you design the control, because the policy is itself a control document. Then pick a lane and paper it - customer release with current tools and documented configuration choices, or bank review staffed and performed without exception. The expensive position is the middle.

Test the Exception, Not the Procedure

Every bank I have advised has a wire policy. The question is whether the policy survives the CEO. Run the tabletop - hypothetically, and never by cloning a real executive’s voice without consent and authorization. An employee gets a document-signature request about a confidential acquisition; a convincing caller claiming to be the CEO hands them to an outside attorney; the attorney wants a new-beneficiary wire before a deadline. Then ask the management team six questions. Where does the employee get a verification number independent of this request? Who can review the payment without broadcasting the supposed deal? Can payment authority be changed by a phone call or an email – any phone call, from anyone? Does a second person or system independently verify the destination before release? Can the CEO waive the process – and if the answer is yes, you have found the vulnerability, because so can anyone who sounds like the CEO. And is the employee explicitly backed, in writing, for delaying a payment pending verification? Grade the exercise on one standard only: was the payment stopped or properly escalated. Whether anyone guessed the voice was synthetic is beside the point – that is a test of hearing, not of controls.

My final comment: prepare, too, for the payment that gets through - immediate contact with the receiving institution, a recall request, an IC3 complaint with full transaction detail, records preserved - while promising yourself nothing, because recovery after release is a hope, not a control. The deeper lesson of this indictment is the one the stopped wire teaches: the bank does not have to win every argument with a convincing impersonator. It must keep the impersonator from ever becoming the payment authority. A control is only meaningful if it survives pressure from someone who appears powerful and in the age of cloned voices and images, “appears powerful” is now a product feature. Build the process that holds anyway.

CCG Catalyst advises banks, credit unions, and fintech companies on payments strategy, operational risk, and vendor management. If your institution wants to test its payment controls against executive-impersonation scenarios, reach out to our team at www.ccgcatalyst.com.

See our latest announcement: CCG Catalyst’s Paul Schaus Named a 2026 Top Consultant by Consulting Magazine

By: Paul Schaus | Founder & Managing Partner, CCG Catalyst Consulting


Disclaimer: The views expressed in this article represent the perspective of CCG Catalyst Consulting based on our direct experience advising financial institutions. This commentary is intended to stimulate industry discussion and does not constitute legal, accounting, or regulatory advice.

Subscribe to our Insights