What Is an AI Voice Scam?

What Is an AI Voice Scam?

An AI voice scam is a social engineering attack delivered by phone call, voicemail, or meeting app in which the attacker uses AI-generated audio, usually a cloned voice of a specific person, to persuade the target to send money, reset credentials, or grant system access. It is a form of vishing.

Conventional vishing relies on a live caller or a prerecorded script, and an AI voice scam adds synthetic speech that sounds like a colleague, executive, or official the target already trusts. Regulators already cover its parts: vishing may incorporate AI-generated voices (opens in new tab), and since a 2024 ruling the Telephone Consumer Protection Act (opens in new tab) treats cloned voices as "artificial."

How AI voice scams work

A cloned-voice scam starts with a sample of the target's voice. Attackers pull audio from public talks, videos, online meetings, and voicemails. Zero-shot text-to-speech models can clone a voice from a few seconds (opens in new tab) of reference audio, and threat actors have sold their own cloning tools (opens in new tab) on Telegram.

Pre-generated audio is the simpler route, and real-time voice conversion (opens in new tab) lets a live caller sound like the target mid-call. Attackers can spoof caller ID (opens in new tab) over VoIP and add urgency, such as a confidential deal or a locked account.

In early 2025, Italian business leaders received calls that appeared to come from Rome government offices, then heard a cloned voice of Defense Minister Guido Crosetto (opens in new tab) asking for money to free kidnapped journalists. One executive sent almost €1 million before Italian police traced the money to a Dutch account and froze it.

Why AI voice scams are hard to stop

Voice checks fail, and email-centric defenses do not inspect the call.

• Listeners cannot consistently identify (opens in new tab) AI-generated voices, so a familiar voice proves nothing.
• Automated voice checks fail too. A BBC reporter's cloned voice passed voice ID checks (opens in new tab) at two UK banks by playing back "My voice is my password." STIR/SHAKEN caller ID authentication works only on IP networks and does not require providers to verify (opens in new tab) caller-supplied identity.
• Callback-phishing emails carry only a phone number, so an email filter has no link or attachment to inspect before the call begins.
• Attackers also skip email and call employees on Microsoft Teams while posing as IT (opens in new tab). Cross-tenant Teams vishing (opens in new tab) is now a more prominent access vector.
• Legacy security awareness training (SAT) built on email templates exercises the inbox, not the phone or the Teams call.

Types of AI voice scams

Enterprises meet four patterns, sorted by whose identity the caller borrows: an executive, a help desk or IT employee, an official, or a family member.

• Executive impersonation for wire transfers. Criminals reportedly use generative AI to impersonate an executive (opens in new tab) or trusted employee and instruct victims to send large sums to attacker-controlled accounts. Finance staff take the call, and the ask is usually an urgent, confidential wire.
• Help desk and employee impersonation for credential resets. Attackers call a service desk as an employee, or call an employee as IT support, to get a password or multi-factor authentication (MFA) reset, a pattern the MITRE ATT&CK framework classes as spearphishing voice (opens in new tab). The financially motivated group UNC6040 impersonated IT support to get targets to authorize a malicious connected app (opens in new tab), and a cloned voice makes a weakly verified help desk easier to fool.
• Official and public-figure impersonation. Actors have used AI-generated voice messages to pose as senior US officials, moving targets to an encrypted messaging app (opens in new tab) almost immediately to request wire transfers, authentication codes, and personal information. The identity carries the authority, so anyone who would take that person's call is a target.
• Extortion and family-emergency pretexts. Scammers use a short audio clip of a loved one's voice (opens in new tab) to pose as a relative in a crisis and demand immediate payment or a ransom. Executives and their families are the enterprise-relevant targets.

How to defend against AI voice scams

Treat a caller's voice as unverified and verify the request instead. Test every payment, credential reset, and account access request against a channel the caller does not control.

1. Callback on an independently sourced number. Hang up, look up the number yourself, and call back before acting on a payment or access request.
2. Multi-person approval for payments and account changes. Require approval levels that escalate with payment size, and bar any single person from granting an exception on a caller's say-so.
3. Stepped help desk verification. Use stepped verification (opens in new tab) for identity reset and account recovery request, and review how help desk staff authenticate (opens in new tab) employees before granting escalated privileges. Delay or deny high-risk password and MFA resets even when verification adds friction.
4. Phishing-resistant MFA. Use FIDO/WebAuthn or PKI-based (opens in new tab) factors, which a caller cannot talk an employee into reading out, unlike one-time codes and approvals exposed to push bombing.
5. Detection as one signal. Add real-time verification (opens in new tab) and passive detection, and protect high-priority officers' communications. Treat a detector's verdict as one input staff weigh, not a gate that approves a request on its own.
6. Limit public voice exposure. Executives' talks, podcasts, and videos supply cloning samples, so limit clean public audio and flag content (opens in new tab) that purports to come from senior leaders.
7. Simulate the voice channel. Train staff with voice and video impersonation exercises (opens in new tab), which let finance and help desk teams rehearse the calls above.

Layered together, these controls turn a convincing voice into one unverified signal among several.

How Doppel helps

Doppel is the Agentic Social Engineering Defense (SED) platform unifying Digital Risk Protection (DRP) and Human Risk Management (HRM), so threats it detects against your executives become the vishing exercises your employees practice.

Its Executive Protection product detects and dismantles fake profiles, spoofed executive emails, and deepfake content tied to leadership. The Doppel Threat Graph links campaign phone numbers, social handles, and domains into one view, and its agentic AI prioritizes and executes takedowns on the infrastructure it maps. Analysts review novel or escalated cases.

Doppel Simulation runs live voice-call simulations with deepfake or AI-generated voices, including Microsoft Teams meetings, and converts a detected threat into a simulation in one click, so finance and help desk teams rehearse the pretext they will face.

Request a demo to see how executive impersonation takedowns and live vishing simulations prepare your finance and help desk teams.

Frequently asked questions about AI voice scams

What is an AI voice scam?

An AI voice scam is a fraud in which a caller uses a synthetic or cloned voice to make a social engineering (opens in new tab) request sound legitimate. The goal is a payment, a credential, or system access the target would refuse from a stranger. Attackers build the clone from public talks, videos, or voicemails and deliver it by phone, voicemail, or meeting app, often from a spoofed number.

What is an AI voice scam in cybersecurity?

In cybersecurity, an AI voice scam is the live-contact step of a multi-step social engineering attack on an organization. Typical objectives are a fraudulent wire transfer, a password or MFA reset obtained through the help desk, or remote access to an employee's machine. Attackers often chain channels, moving from a message to a call and then to an encrypted app. Cloned audio is synthetic content (opens in new tab), and voice-based help desk attacks map to spearphishing voice (MITRE ATT&CK technique T1566.004).

AI voice scam vs. vishing: what is the difference?

Vishing covers any voice-based social engineering, whether the voice belongs to a human, a recording, or a synthetic source. An AI voice scam is the subset in which the voice is synthetic, usually a clone of someone the target knows or trusts. The difference matters for defense: a familiar voice stops being evidence of identity, so controls must verify the request instead.

What is an example of an AI voice scam?

In 2025, an impersonator used AI to mimic US Secretary of State Marco Rubio's voice (opens in new tab) and writing style, then left voicemails for at least two targets (opens in new tab) and invited another to continue on Signal. The recipients included three foreign ministers, a US governor, and a member of Congress. The case shows the pattern: a trusted identity, an unexpected channel, and a push to move the exchange somewhere harder to verify.