AI voice cloning and deepfake BEC: how the attack works and how to verify requests

For years the advice for a suspicious payment email was simple: pick up the phone. Voice cloning and deepfake video weaken that advice, because the voice on the phone or the face on the screen can now be generated. This guide explains how criminals combine email with synthetic voice and video to impersonate executives, what US agencies have actually reported, and the verification procedures that still work. It builds on our guides to AI phishing emails and business email compromise.

What is deepfake BEC?

Business email compromise (BEC) is fraud where a criminal poses as a boss, colleague or supplier to get money or data. Deepfake BEC adds AI-generated audio or video to make the request more convincing.

US agencies describe the building blocks:

• Cloned voices. The FBI's December 2024 PSA says "Criminals can use AI-generated audio to impersonate well-known, public figures or personal relations to elicit payments," and that criminals "obtain access to bank accounts using AI-generated audio clips of individuals and impersonating them."
• Live video. The same PSA says criminals "generate videos for real time video chats with alleged company executives, law enforcement, or other authority figures."
• Executive impersonation for transfers. FinCEN's November 2024 alert says criminals "have reportedly used GenAI tools to target companies by impersonating an executive or other trusted employee and then instructing victims to transfer large sums or make payments to accounts ultimately under the scammer's control." A footnote to that sentence cites a February 2024 CNN report headlined "Finance worker pays out $25 million after video call with deepfake 'chief financial officer'." FinCEN cites the report as a source; the alert doesn't describe the case itself.
• Email plus voice. The FBI's 2025 IC3 Annual Report says "chat generators can quickly create official-sounding emails mimicking a company's CEO" and "voice cloning can also be used to request wire payment." Businesses reported losses over $30 million to BEC involving AI in 2025.

The reported AI figure is small next to total BEC losses of $3,046,598,558 in the same report, and the FBI notes "not all are AI-enabled." But the reported figure only counts cases where victims recognized AI, and the techniques are getting cheaper.

How does a deepfake BEC attack unfold?

Most attacks combine channels. Email does the setup; voice or video does the persuasion.

Step 4 shows up repeatedly in FBI warnings. In the campaign impersonating senior US officials, first reported in May 2025, actors sent "AI-generated voice messages" and then sent "a malicious link under the guise of transitioning to a separate messaging platform." The FBI's December 2025 update says the activity dates back to 2023, that contact usually starts by text, and that actors quickly move victims to Signal, Telegram or WhatsApp. There they ask for authentication codes, copies of documents such as passports, wire transfers to overseas banks, and introductions to other contacts.

That campaign targeted officials and their contacts, not finance teams, but the playbook transfers directly to business: impersonate someone senior, use a cloned voice to build trust, move to an unmonitored channel, then ask.

Deepfakes in hiring

A related risk is the fake candidate. The 2025 IC3 report describes voice spoofing, "or potentially voice deepfakes," in online job interviews, where lip movement doesn't match the audio and coughs or sneezes aren't aligned with the video. The FBI says the goal "generally appears to be gaining access to private computer networks." If you hire remotely, make identity checks part of onboarding.

Can you spot a cloned voice or deepfake video?

Sometimes. Agencies list tells to watch for:

• Voice. The FBI suggests listening closely to "the tone and word choice," and its May 2025 warning lists "voice call lag time" as a sign of a fake. The FBI and ABA Foundation infographic asks: "Is the voice tone flat or unnatural?"
• Video. The infographic asks whether facial features are "blurry or distorted," whether the person blinks "too much or too little," whether "the audio and video [are] out of sync," and whether there are "odd or unnatural shadows or lighting." The December 2024 PSA adds lag time, voice matching and unrealistic movements.
• Behavior. FinCEN notes that people using deepfakes may try to avoid live checks, for example by claiming "repeated technical glitches" or asking "to change communication methods" during verification.

Don't build your defense on detection. The FBI's December 2025 update warns that AI-generated voices "can sound nearly identical" to the people they imitate. The infographic's red flags are more reliable because they're about the request, not the media: "Emotional manipulation involving fear or urgency," "Unexpected requests for money, passwords, personal information or secrecy," and "Uncharacteristic communication from someone you know, especially over text, phone or video."

What verification procedures stop deepfake BEC?

The aim is a process that works even if the voice and face are perfect. Verification has to happen on a channel the criminal doesn't control.

1. Call back on a number you already hold

The FBI's advice is to "verify the identity of the person calling you by hanging up the phone, researching the contact of the bank or organization purporting to call you, and call the phone number directly." In a business:

1. End the call or meeting politely, whoever it appears to be.
2. Look up the person's number in your own directory, contract file or HR system. Never use a number, link or meeting invite from the request.
3. Call that number and confirm the request in your own words.
4. Record who you called, when, on which number, and the answer.

2. Agree a code phrase for urgent requests

The FBI recommends families "create a secret word or phrase" to verify identity, and its 2025 warnings repeat it. The business version is a challenge phrase agreed in person between executives and the people who act on their requests. Keep it off email and chat, change it when someone leaves, and treat it as an extra check, not a substitute for the callback.

3. Require two people for money and access

No single person should be able to complete a payment or bank-detail change on instruction alone, however senior the requester appears. Require a second approver for transfers above a threshold, any first payment to a new account, and any change to supplier or payroll bank details. Our vendor invoice fraud guide has a ready-made procedure.

4. Keep approvals inside company systems

Make it a written rule that payment approvals, password resets and data requests never happen on personal messaging apps or in unscheduled video calls. A request to switch channels is a red flag in itself.

5. Protect the help desk

Cloned voices are used against help desks as well as finance. Before resetting a password or MFA method, verify the caller through a callback to the number on file or a check by their manager. Never accept a voice on the phone as the only proof. Never give out MFA codes. The FBI says: "Never provide a two-factor code to anyone over email, SMS/MMS text message, or encrypted messaging application."

6. Make it safe to say no

Executives should tell staff, in advance, that they will never ask for a wire, gift cards or secrecy on a call, and that anyone who delays a request to verify it will be thanked. Pressure to skip the process is the attack.

Which email controls still matter?

Deepfakes get the headlines, but most of these attacks still depend on an email at step 2. Blocking it means the call never happens.

Check your domain's policy with the DMARC Checker, and see how to stop spoofing of your domain. For the impersonation tricks DMARC doesn't cover, read our guides to display name spoofing and lookalike domain phishing. If a message looks off, the Email Header Analyzer shows where it really came from.

What should you do if you suspect a deepfake?

1. Pause the request. Don't pay, reset or send anything.
2. Verify out of band using the callback steps above.
3. Preserve evidence. Keep the email with full headers, voice notes, call logs, meeting links and chat messages.
4. If money moved, call your bank immediately and ask for a recall. Then file a complaint at ic3.gov with the transaction details. The infographic also advises reporting to your bank and local police.
5. Warn the person who was impersonated, and check their mailbox for signs of compromise using our guide to the signs an email account is compromised.
6. Tell staff that a campaign is under way, without blame.

Next steps

1. Write a one-page verification rule covering callbacks, code phrases, dual approval and no approvals on personal apps. Have every executive endorse it.
2. Run a tabletop exercise: an email from the CEO plus a voice note asking for an urgent transfer. See who follows the rule.
3. Add deepfake examples to your security awareness training.
4. Check your domain's SPF, DKIM and DMARC with the free Email Health Check.
5. Put filtering in front of your mailboxes that catches spoofed senders and impersonation before the setup email lands. iBlockSpam filters phishing, spoofing and malware at the MX for Microsoft 365, Google Workspace and cPanel; see how it works.

Browse our other phishing and threat guides.