← Back to articles
voice deepfakeceo fraudvoice cloningsocial engineeringcybersecurity

Voice deepfake of your boss: how to verify a call is real

By · November 10, 2026 · 6 min read

A voice deepfake uses AI to imitate a specific person, such as your boss, to ask you for an urgent payment. Don't try to judge it by ear: hang up, call back on the number you already have, and make sure no payment can be approved by phone alone.

File JG · DocumentBack to articles

Voice deepfake of your boss: how to verify a call is real

Type
Article
Author
Date
Reading time
6 min read
Subjects
voice deepfake, ceo fraud, voice cloning, social engineering, cybersecurity

Summary

A voice deepfake uses AI to imitate a specific person, such as your boss, to ask you for an urgent payment. Don't try to judge it by ear: hang up, call back on the number you already have, and make sure no payment can be approved by phone alone.

Your phone rings and it's your boss's voice. She's in a hurry, about to walk into a meeting, and needs you to make a payment today. The tone is hers, and so are the little verbal habits. A few years ago, recognising the voice was enough to feel safe. Not anymore: a voice can now be cloned with artificial intelligence, and the trick has become a version of CEO fraud that every company should have on its radar.

In this article I explain what a voice deepfake is, how it has already been used against a company in Spain and, above all, how to check a suspicious call without relying on your ears.

◢What is a voice deepfake?

It's a fake voice generated with artificial intelligence that imitates a specific person. You'll also see it called voice cloning, an AI-cloned voice or deepvoice. INCIBE1, Spain's National Cybersecurity Institute, files it under voice hacking, the techniques used to manipulate people's voices, and describes it as something that produces fake messages that seem to come from a real person.

What matters for a business is how it gets used. A cloned voice lends credibility to a request that would already look suspicious by email: an urgent transfer, a change of bank details, a set of login credentials.

◢Has it already happened?

Yes. In January 2024 INCIBE published a real case handled by its cybersecurity helpline2. An employee got a call in which the voice was exactly like the CEO's, asking for a phone meeting with several members of staff. In that meeting, the fake CEO told them they needed to buy a machine and gave them instructions for a bank transfer. They made it. They only realised it was a scam days later, when they spoke to the real CEO again.

Notice that nobody did anything foolish. They got an instruction from someone with their boss's voice, with a believable reason, and they followed it. It's the same mechanism I describe in CEO fraud: how to spot it, with one difference: the check many people relied on, "I heard him, it was him", no longer works.

◢Can you tell a cloned voice from a real one?

Sometimes, but you shouldn't count on it. In a May 2025 alert3 about AI-generated voice messages impersonating senior US officials, the FBI recommends listening closely to tone and word choice, but admits that a legitimate call and a cloned voice can sound nearly identical, and that AI-generated content is often difficult to identify.

In the same alert, the FBI mentions signs such as lag on the call or a voice that doesn't quite match. They're useful clues, but if your company's plan is for the finance team to spot the fake by ear, the plan depends on luck.

That's why the right approach is different: don't try to guess whether the voice is real, verify the request through a route the attacker doesn't control.

◢How to check whether a call from your boss is real

These are the checks that work. Almost all of them come from INCIBE's and the FBI's advice:

  • ▹Hang up and call back yourself. The FBI advises finding the person's number independently and calling them to verify. INCIBE says to always double-check through other channels you already use with that person. Never call back the number that called you, or one you're given during the conversation.
  • ▹Be wary of a change of channel. If someone you know contacts you from a new number or app, the FBI recommends confirming that contact through a channel you had already verified.
  • ▹Ask something only the real person would know. It doesn't need to be an interrogation: a detail from your last meeting or an internal project is enough to throw an impostor off.
  • ▹Look at the context, not just the voice. The FBI stresses judging whether the request makes sense. An urgent, confidential payment outside the usual process is suspicious however perfect the voice sounds.
  • ▹Never give codes or passwords over the phone. The FBI warns that scammers may try to talk you into reading out a two-factor code so they can take over the account.

None of these checks needs any technology. What they need is for the person taking the call to be allowed to hang up on the boss without consequences.

◢What your company can do so it doesn't work

The best defence is making sure that a phone call, however convincing, is never enough to move money. In the fake CEO case, INCIBE recommends that payments be authorised by email or in person, never by phone, and that large amounts be approved by more than one person.

On top of that, I would add:

  • ▹Write down the procedure for payments and bank detail changes, with no exceptions "because management asked".
  • ▹Agree a verification word or phrase among the people who authorise payments. The FBI recommends families set up a secret word to confirm identities, and the same idea works in a small team. Don't keep it in email or the company chat.
  • ▹Limit what gets published about management's schedules and travel. In a December 2024 alert4, the FBI advises limiting public content featuring your voice and image to make it harder for fraudsters to build fake identities.
  • ▹Tell the whole team when there's an attempt, as INCIBE recommends, so the next one doesn't catch anyone off guard.

If you want the procedure to hold up against emails too, here's how to tell if an email is phishing.

◢What if you've already made the payment?

In the real case above, INCIBE recommends contacting the bank straight away to try to cancel the transfer and, if it doesn't respond, taking the case to the Bank of Spain. It also advises blocking the number, warning all staff, keeping the evidence (call logs, numbers, recordings) and reporting it to the police. The executive who was impersonated can report it too.

In Spain, INCIBE's cybersecurity helpline answers on 017, which is where the company in that case turned. Elsewhere, go to your bank and your national police or cybercrime reporting service; in the US, the FBI asks victims to report to IC35.

And one human detail: whoever made the payment often takes a while to say so out of embarrassment. The sooner it's known, the better the chance of stopping the money.

◢Can you train your team for this?

You can, and it's where you gain the most. Reading that cloned voices exist doesn't change what you do when the phone rings and the boss is in a hurry. What changes the reaction is having practised beforehand: I hang up, I call the number I already have, and that's fine.

It's one of the scenarios I work through in my cybersecurity and AI training for companies, with examples designed for non-technical teams.

Related documents

  1. News

    Claude Fable 5.2 and Mythos 5.2: what's new from Anthropic

    Anthropic launches Claude Fable 5.2, open to everyone with cybersecurity and biology safeguards, and Mythos 5.2, the same model without those brakes for vetted organizations. What improves, the cybersecurity part, where it is and what it costs.

  2. News

    mittwald hosting breached through its MCP server for AI

    German hosting provider mittwald suffered an attack that got in through the MCP server its customers used to connect AI assistants. Session tokens were stolen and 56 projects compromised. What we know and what to check if you use AI connectors.

  3. News

    OpenSSH SSHole flaw exploited: what to check if you run Alpine Linux

    CISA confirms SSHole (CVE-2026-123158) is being exploited: an OpenSSH flaw that lets attackers log in with a made-up password to password-disabled accounts on Alpine Linux and other musl systems. Fixed in OpenSSH 10.6p2. Who is affected and what to do.

End of documentBack to articles