The call comes on a Friday afternoon. It is your managing partner’s voice, unmistakably, right down to the way she clears her throat before a number. She is closing something confidential, she needs a payment reissued to a new account, and she needs it before the wire cutoff. The caller ID checks out. The voice checks out. Everything checks out except the one thing that matters: it is not her.

Vishing, voice phishing, is the phone version of the fraud emails you already ignore. It used to be easy to dismiss because a human had to do the talking, and humans doing scam calls at scale sounded like scam calls. AI voice cloning removed that limit. This guide covers what changed, what the calls look like, and the procedures that stop them, none of which require buying anything.

Why the phone works when email fails

Your email has defenses working quietly on your behalf: filters, quarantine, and the authentication records that make impersonating your domain hard. A phone call has none of that. There is no spam folder for a live conversation, no banner warning that the caller is external, and no time to reflect, because the person on the line is applying pressure in real time.

That is why the phone is having a moment. CrowdStrike’s 2025 threat report documented a 442% increase in vishing attacks over the second half of 2024, and Mandiant found voice phishing had become one of the most common ways attackers get their initial foothold. The growth is not because criminals discovered telephones. It is because the two hard parts of a convincing call, a credible voice and a plausible story, both got automated.

What AI actually changed

Cloning a voice used to require studio time and skill. It now requires a short audio sample and a consumer tool. The sample is the easy part: a voicemail greeting, a webinar, a podcast interview, the founder video on your own website. If someone at your company has ever spoken into a microphone, their voice is available.

Two other pieces complete the kit. Caller ID spoofing makes the call appear to come from a number you recognize, and it has been trivial for years, because caller ID was never designed to prove anything. And the same email compromises that fuel invoice fraud now supply the script: an attacker who has read a mailbox knows the deal names, the vendor relationships, and the internal shorthand that make a call sound like it could only be internal.

Three calls that actually happened

Arup, 2024. A finance employee in the engineering firm’s Hong Kong office received an email from the UK-based CFO about a confidential transaction, then joined a video call with the CFO and several colleagues. Every other person on that call was an AI-generated fake. The employee made fifteen transfers totaling about $25 million. The fraud surfaced only when the employee later checked with head office.

Ferrari, 2024. An executive received WhatsApp messages, then a call, from someone using CEO Benedetto Vigna’s cloned voice, southern Italian accent included, about a confidential acquisition needing discretion and an urgent signature. The executive noticed something slightly off in the intonation and asked a question only the real Vigna could answer: the title of the book he had recommended days earlier. The caller hung up. That is the entire cost of a working verification habit, one question.

MGM Resorts, 2023. Attackers found an employee’s details on LinkedIn, phoned the IT help desk impersonating them, and talked their way into a credential reset. No cloned voice was needed, just confidence. The resulting breach disrupted hotel operations for days and cost the company roughly $100 million by its own filing. The help desk call is worth as much to an attacker as the CEO call, because a password reset opens more doors than a wire transfer.

The playbook is always the same

Strip away the technology and every one of these calls runs on three ingredients: authority (the boss, the CFO, the bank’s fraud department), urgency (the wire cutoff, the account under attack, the deal closing today), and secrecy (confidential, do not loop anyone in, I am trusting you with this). Secrecy is the tell that does the most work, because its only real function is to keep you away from the person who would say “that’s not me.”

The targets are predictable too. Finance staff get payment and bank-detail changes. IT help desks get password and MFA reset requests. Everyone else gets asked for the small things that add up: a gift card purchase, a code that just arrived by text, a quick confirmation of someone’s mobile number.

Procedures beat instinct

The Ferrari executive got lucky in one sense: they happened to be suspicious. A procedure means nobody has to be. The defenses below cost nothing and do not depend on anyone detecting a fake, which matters because the fakes are now good enough that detection is not a plan.

  • Verify on a number you already have. Any request to move money, change bank details, or reset credentials gets confirmed by calling the requester back on the number in your directory, never the number that called you and never a number the caller helpfully provides. This single habit defeats the voice, the caller ID, and the script all at once.
  • Put payment changes in writing, as a rule with no exceptions. Bank-detail changes and unusual payments require callback verification regardless of who asks, how senior they sound, or how urgent it is. The no-exceptions part is the load-bearing part: the moment seniority or urgency can waive the rule, the rule selects for exactly the calls attackers make. Writing it down is what makes it a rule rather than a habit, which is the difference between a policy and a good intention.
  • Treat secrecy as the alarm. Legitimate confidential deals still survive one verification call. Train the team to hear “don’t tell anyone” as the start of the fraud, not the start of the favor.
  • Never read a code to anyone. No bank, no IT department, and no colleague has a legitimate reason to ask for an MFA code over the phone. Codes sent by text are weak enough that Microsoft is already retiring them; reading one to a caller defeats even the good kind.
  • Give the help desk a verification script. Whoever resets passwords for your business, in-house or outsourced, should verify identity through something stronger than name plus date of birth: a callback to the number on file, a manager confirmation, or an in-person check for sensitive accounts. Ask your IT provider what their reset procedure is. The answer is revealing.

A shared verification phrase for the leadership team, agreed in person and never written into email, is a reasonable extra layer for approving urgent payments. Just treat it as a supplement to callbacks, not a replacement, since anything spoken on a compromised channel eventually leaks.

What a prepared business looks like

  1. A written payment-change procedure requiring callback verification on a known number, with no seniority exception.
  2. A help desk that verifies identity before resetting passwords or MFA, every time.
  3. Staff who treat urgency plus secrecy as the signature of fraud, not of importance.
  4. Nobody reads codes over the phone. Ever, to anyone.
  5. Verification numbers kept in a directory, not taken from the incoming call or message.
  6. The finance team drilled on this specifically, because they are the ones the calls are written for.

Notice what is absent: nothing here detects deepfakes, and nothing costs money. The entire defense is refusing to let a voice, however convincing, be sufficient authorization on its own. That holds whether the voice is cloned or the real thing under duress.

Where to start

Ask two questions this week. If the CEO’s voice called your bookkeeper today asking for an urgent transfer to a new account, what exactly would happen next? And if someone called your IT support claiming to be you and asked for a password reset, what would they be asked to prove? If either answer is “it would depend on who picked up,” that is the gap.

Verification procedures and the training to make them stick are part of our cybersecurity service, including simulations that test whether the procedure survives contact with a convincing caller. Get in touch and we will help you close the phone-shaped hole in your defenses.