The phone rings. It's your boss's voice, slightly stressed: a supplier deal is about to fall through, and a payment needs to go out in the next hour. They're boarding a plane and can't talk long. The voice is perfect, because it was generated by AI from a podcast interview.

Deepfake scams, which use AI to clone voices and even faces on video calls, have moved from rare headlines to a common risk for businesses of every size. The good news: simple procedures stop almost all of them.

How the scams work

ScamHow it works
CEO voice fraudA cloned voice of the owner or a director calls staff demanding an urgent payment
Fake video callA video meeting with deepfaked colleagues approving a transfer
Supplier impersonationA cloned voice or AI-written email from a "supplier" announcing new bank details
Fake customer or job candidateDeepfaked video in interviews to gain access or steal data
Account takeoverCloned voices used to pass phone-based identity checks at banks or service providers
Family emergency scamsTargeting staff personally, a common way into business accounts

Attackers gather material from websites, social media, podcasts, webinars and even voicemail greetings. AI-written emails without spelling mistakes make the old "look for typos" advice useless.

Warning signs

  • Urgency and secrecy: "It has to go now", "Don't tell anyone yet".
  • Unusual channels: a WhatsApp voice note or a new number instead of normal contact.
  • Changes to payment details, especially just before a payment is due.
  • Pressure to bypass normal approval.
  • Odd timing: late evenings, holidays, when the "caller" is supposedly travelling.
  • Slight oddities: unnatural pauses, flat emotion, background noise that doesn't fit, video glitches around the face.

Don't rely on spotting glitches. The technology improves constantly; procedures beat perception.

Seven rules that stop deepfake fraud

1. Verify through a separate channel

Any request to pay money, change bank details or share sensitive data must be confirmed by contacting the person through a known number or channel you already have, never the one the request came from.

2. Use a code word

Agree a private code word or question among people who approve payments. Ask for it when anything feels off.

3. Two-person approval for payments

Payments above a set amount need two people to approve. No exceptions for urgency.

4. Lock down bank detail changes

Supplier bank changes require written confirmation, a call-back to a known number and a waiting period before the first payment.

5. Train your team

Short, regular training with real examples. Make it clear that staff will never get in trouble for slowing down a payment to verify it, even if the request really was from the boss.

6. Reduce what attackers can use

You can't hide your voice entirely, but you can:

  • Avoid publishing long, clean voice recordings of key finance approvers where it isn't needed.
  • Use generic voicemail greetings.
  • Limit what staff share publicly about internal processes and travel.

7. Strengthen accounts

Use phishing-resistant sign-in (passkeys or hardware keys) for email, banking and admin accounts, so stolen passwords or voice-based tricks don't open everything. Read passkeys explained.

A simple payment verification policy

  1. No payment or bank detail change is made on the basis of a phone call, voice note, video call or email alone.
  2. All such requests are verified by calling the requester on a number from our records.
  3. Payments over £[amount] need approval from two people.
  4. New or changed supplier bank details are confirmed in writing and by call-back, and the first payment waits 24 hours.
  5. Anyone can pause a payment to verify it, without blame.

Print it, share it and make it part of onboarding.

If you think you've been targeted

  1. Stop: don't make further payments.
  2. Call your bank immediately if money was sent; quick action improves the chance of recovery.
  3. Report it: in the UK to Action Fraud (or Police Scotland), and to the relevant authority elsewhere.
  4. Secure accounts: change passwords, check email forwarding rules and sign-in logs.
  5. Tell the team so others aren't caught.
  6. Review what happened and update your procedures.

Bigger picture: AI-powered threats

Deepfakes are part of a wider shift: AI-written phishing, automated attacks and cloned identities. Basic security hygiene still blocks most of it: supported systems (see Windows 10 after end of support), updates, strong authentication, backups and trained staff. Read website security basics for small business.

The short version

AI can clone voices and faces convincingly, and criminals use them to impersonate owners, directors and suppliers. Don't rely on spotting fakes. Verify every payment and bank detail change through a separate, known channel, use code words and two-person approval, train staff to pause without blame and secure accounts with passkeys.

Want a security check-up for your business systems? Our maintenance and support team can help.