AI Vishing and Voice Cloning: How Employees Can Verify the Caller

Imagine receiving an urgent call that sounds like your manager. The caller knows the project name, mentions a real colleague and asks you to approve a sensitive action immediately. Nothing about the conversation feels unusual—except the request itself.

This is the risk behind AI vishing. Attackers can use synthetic voices, cloned audio or AI-assisted conversational scripts to make fraudulent calls more convincing. Instead of teaching people to identify an artificial voice by sound alone, organisations need reliable verification procedures.

What is AI vishing?

Vishing means voice phishing: using a phone or voice conversation to deceive someone into handing over information or taking an unsafe action. AI vishing introduces AI-generated scripts, synthetic speech or voice impersonation into that process.

Some attacks imitate an individual voice. Others use a convincing but entirely invented helpdesk or banking persona. Both can exploit a recipient's instinct to trust a confident caller.

What an attack can look like

A common scenario begins with information gathering. The attacker learns who approves payments, manages employee accounts or supports senior leadership. They then create a plausible reason for a call, often involving a deadline or a supposed security incident.

During the conversation, the caller may request an MFA code, a password reset, a remote-access tool installation, changes to supplier details or an urgent payment. The apparent familiarity of the speaker can distract from the fact that the requested action breaks normal controls.

A phone call can also follow a phishing email or SMS. That second contact may appear to confirm the first request when both are controlled by the same attacker. PhishSheriff's multi-channel simulation overview describes why verification should remain consistent across channels.

The verification rule: trust the process, not the voice

Employees should be authorised to interrupt a conversation when something is unusual. A simple response is: I will confirm this using our normal process and get back to you.

Then:

1. 1End the call if necessary. Do not use a number supplied by the caller.
2. 2Contact the purported requester using an existing directory entry or known company number.
3. 3Apply approval rules for payments, account recovery, bank changes and privileged access without exception.
4. 4Refuse requests for passwords or one-time authentication codes.
5. 5Report the interaction promptly, including the number, time and request made.

For finance and HR functions, a second authorised approver can provide a particularly useful control against urgent requests to override normal practice.

What security teams should rehearse

One training module cannot cover every possible voice scam. Choose scenarios with clear business relevance: an IT caller requesting an MFA code, a supplier urging a bank-detail update, a senior executive seeking confidential data or a caller escalating an invoice request that began by email.

Make the expected behaviour explicit. Did the person verify identity independently? Did they refuse an unsafe request? Did they escalate? These observations are more useful than simply counting answered calls.

PhishSheriff's AIVA voice phishing simulation supports controlled voice-based exercises and follow-up training. These exercises should be clearly authorised, proportionate, and aligned with company policy.

Build a verification culture

People are more likely to follow secure procedures when management visibly supports them. Senior leaders should tell staff that an urgent instruction does not bypass verification, and nobody will be penalised for checking an unusual request.

For a broader exercise involving finance, security and leadership, use the cybersecurity tabletop exercises guide to plan how a suspicious call becomes a reported incident.

Frequently asked questions

Can people reliably recognise a cloned voice?

Not reliably enough to make voice recognition the primary control. Confirm sensitive requests through a separate, trusted process.

Is all vishing AI-powered?

No. Traditional human callers can be just as deceptive; both require the same verification habits.

What should an employee do after sharing an MFA code?

Immediately contact the security or IT response team using the approved channel. Prompt reporting helps the team secure the account and investigate activity.