What is vishing? The AI voice scam that clones you in 3 seconds

You get a call. It's your boss's voice — the tone, the cadence, the accent, everything matches. He asks you to transfer $50,000 urgently to close a contract. You hesitate for two seconds and make the transfer. Only hours later do you find out the "boss" was an AI that cloned his voice from a 3-second clip pulled out of a recorded meeting.

That isn't a TV script. It's vishing — and attacks grew 442% between the first and second half of 2024, according to CrowdStrike's Global Threat Report 2025. In the first half of 2025 the volume had already surpassed all of 2024.

What is vishing?

What does vishing mean?

Vishing is short for voice phishing — phishing carried out over a voice call, whether by landline, mobile or VoIP. Instead of an email with a malicious link, the attacker calls pretending to be someone trustworthy: a bank, tech support, the tax authority, a relative or your boss. The goal is the same as any social engineering: extract credentials, one-time passwords (OTP) or convince the victim to transfer money.

What has changed over the past two years is the tooling: AI voice cloning models turned the attack surgical. A three-second clip — from a public video, a podcast or a recorded meeting — is already enough to generate a convincing replica of anyone's voice.

How AI transformed vishing

Vishing has been around since the 1990s, when scammers used auto-dialers to pretend to be banks. What sets today apart is voice cloning. Commercially available tools — and clandestine versions on criminal forums — let anyone:

• Extract an audio snippet from a YouTube video, an interview or a recorded meeting;
• Feed it into a voice synthesis model;
• Generate synthetic audio in real time during the call.

In Q1 2025, vishing attacks grew 1,633% over the previous quarter, according to CrowdStrike data. C-suite executives are the most frequent target — but anyone with a public presence online (interviews, podcasts, corporate videos) has already supplied enough material to be cloned.

How a vishing attack works in practice

The typical script has five stages:

1. Reconnaissance (OSINT): the attacker gathers public information — LinkedIn, social media, the corporate site — to identify hierarchy, internal jargon and relationships between people.
2. Voice collection: extracts audio of the person to be imitated (usually a superior or a relative) from public sources.
3. Scenario prep: builds a pretext with urgency and authority — "the audit starts tomorrow", "my phone broke, I need access now".
4. Executing the call: uses VoIP with caller ID spoofing to make the call appear to come from a legitimate extension or number.
5. Action: the victim transfers money, hands over an OTP code or installs "remote support" software — which is a RAT (Remote Access Trojan).

Urgency is the central trigger. It stops the victim from pausing, verifying and consulting other people — exactly as happens in the MFA fatigue attack, where a burst of notifications trains the user to approve out of exhaustion.

Confirmed real cases

In February 2025, scammers cloned the voice of Italian defense minister Guido Crosetto and called business leaders asking for an "urgent ransom for kidnapped journalists". The target list included Giorgio Armani and Patrizio Bertelli, but the one who fell for it was businessman Massimo Moratti: around €1 million transferred to accounts in the Netherlands and Hong Kong. Italian police managed to freeze the funds afterwards.

In April 2026, the ShinyHunters group used vishing to compromise Charter Communications. An employee was talked into providing access to the company's Microsoft Entra account over the phone — from where the attackers reached the Salesforce instance and exfiltrated data from 4.9 million accounts. The same group is behind other large-scale breaches, such as the Canvas LMS incident with 275 million student records.

The most common vector varies by region. In Brazil, it's still fake banking tech support — the voice imitates the tone of a real agent — and phone-call variants of the "hi mum" scam. If you haven't read about that mechanic yet, the post on how to protect yourself from the "hi mum" WhatsApp scam covers the social engineering behind it well.

How to detect a vishing attempt

How do I know if a call is vishing?

Any call that combines urgency, a request for sensitive data and pressure not to hang up is suspicious — regardless of whether the voice sounds familiar.

Other warning signs:

• A call from an unknown number claiming to be someone close to you;
• Audio quality that's slightly robotic or has atypical pauses;
• A request for a code sent by SMS or an authenticator app — legitimate banks never ask for those codes over the phone;
• Pressure not to confirm through another channel ("don't talk to anyone about this right now");
• A request to install an app ("so I can help you better, install this application").

How to protect yourself from vishing

Concrete actions, without relying on instinct:

• Agree on a code word with family and close colleagues. If someone calls in an urgent situation, ask for the word. A cloned AI doesn't know it.
• Hang up and call back on the official number. Never make a decision on the same call. End it, find the number on the official site and call back.
• Implement second-channel verification in financial processes. Any transfer above a defined amount requires confirmation by another means (email, message or in person).
• Never give out an OTP over the phone. No bank, carrier or legitimate company asks for an authentication code by call.
• Lock number portability on your line. Many vishing attacks lay the groundwork for a subsequent SIM swap — ask your carrier for a port-out lock that can't be lifted without showing up in person.

In a corporate setting, limiting public exposure of internal information makes the initial reconnaissance harder. The guide to OPSEC and digital privacy covers that layer in more depth.

The bigger lesson

Vishing exploits a fundamental human limit: we recognize a voice as proof of identity. AI broke that premise. The defense isn't technological — it's procedural. Second-channel verification, a code word agreed in advance and a culture of "can I confirm before acting?" are more effective than any call filter.

The urgency the attacker injects into the call is precisely the signal to slow down. If the situation "can't wait 5 minutes to be verified", it's almost certainly a scam.