The voice can be familiar and still be fraudulent
A familiar voice has traditionally carried a useful operational signal: it may be a senior executive, a long-standing supplier, or a colleague known to the person answering the phone. That signal is now weaker. Voice cloning can make an unexpected request sound consistent with the relationship, while urgency and authority discourage the recipient from asking a second question.
Recent practitioner guidance describes the same practical risk across executive impersonation, supplier fraud and payment requests. The issue is not whether an employee can identify an artificial voice in isolation. It is whether the firm’s workflow treats a voice as evidence of identity when the request involves money, credentials or sensitive information. CIAT’s overview of deepfake social engineering and IntellectIT’s defence guidance both point towards the same conclusion: verification must be independent of the communication channel being used.
The risk is not hypothetical. In early 2024, a finance employee at engineering firm Arup wired about 25 million US dollars after a video call involving AI-deepfaked recreations of the CFO and colleagues, as reported by CNN Business and referenced by the World Economic Forum, Deloitte and FinCEN. Arup’s CIO described the incident as technology-enhanced social engineering. The initial contact was a phishing email, and no systems were compromised.
What readiness looks like in a real workflow
A policy can require call-backs and dual approval. That does not prove the control works at the point of pressure. A readiness test should examine what people actually do when a purported executive asks for an urgent payment, when a supplier requests a beneficiary change, or when a VIP servicing call seeks a password reset.
The DeepBlocker Vishing Readiness Assessment is a fixed-scope, authorised engagement lasting two to three weeks. We design controlled AI red-team calls around the workflows that matter to the firm, then measure whether staff:
- challenge an unusual request without relying on familiarity;
- use an approved, independent channel to verify the caller;
- pause a transaction when the caller creates urgency or invokes senior authority;
- record the event and escalate it to the right owner; and
- continue to follow the process when the caller changes tactics.
This is not a test of whether a person can win a debate with an AI voice. It is a test of whether the operating model creates a safe path to say no, verify independently and obtain timely support.
FinCEN Alert FIN-2024-Alert004, issued on 13 November 2024, warns financial institutions about executive-impersonation vishing and notes that synthetic audio or video can also respond to live prompts. The alert is advisory guidance, not a binding rule. Its practical implication is important: live verification alone is not a complete control.
The evidence a board can use
At the end of the assessment, DeepBlocker provides a board-ready evidence pack. It records the scenarios tested, the workflows involved, who was fooled or correctly challenged, where escalation failed, and which control changes should be prioritised. It gives the MLRO, Compliance Officer, Head of Risk, COO or CISO evidence about behaviour and process, rather than a generic warning about deepfakes.
That evidence can support remediation, insurer discussions and regulatory examinations. DORA Article 26 provides a recurring rationale for testing people, processes and technologies through threat-led penetration testing, including defined boundaries for social engineering. In the UK, the Payment Systems Regulator’s reimbursement regime has applied since 7 October 2024, with reimbursement of up to £85,000 per Faster Payments scam claim. Prevention therefore belongs in the control framework, not only in staff awareness material.
Book an assessment to test whether your high-risk voice workflows survive pressure.