Skip to main content
Neural Defend

How Do Deepfakes Enable Business Email Compromise?

Deepfakes enable business email compromise by adding a cloned voice call, voice note or fake video meeting that seems to confirm a fraudulent email instruction. The email, often from a compromised or spoofed executive or vendor account, asks for a payment or bank-detail change; the synthetic call removes the target's doubt. BEC with deepfake calls is stopped by verifying every payment change through a known channel, whatever the caller sounds like.

Organization authorNeural Defend Editorial
Updated May 26, 2026
Sumit Singh, Co-founder and CDO of Neural Defend.
Sumit SinghCo-founder & CDO
Email, audio, meeting, and document artifacts converge on a payment request that is blocked and traced to compromised access.

How do deepfakes enable business email compromise?

Deepfakes enable business email compromise by supplying the corroboration that well-trained finance staff have learned to ask for. A decade of awareness training taught people to distrust an unexpected email asking for an urgent wire. Attackers responded by adding a second channel: a short voice note in the CFO's voice, a phone call from the "vendor", or a video meeting where the executive appears briefly before claiming a bad connection.

The email still does most of the work. A compromised mailbox gives the attacker real invoice history, names, writing style and timing, and sometimes the ability to reply inside an existing thread. The synthetic media is short and only has to survive a few seconds of scrutiny, often over a compressed call. Both channels are controlled by the attacker, so the victim's attempt to verify actually confirms the fraud.

The FBI Internet Crime Complaint Center documented this shift early. A public service announcement of 16 February 2022 described criminals using a compromised CEO or CFO email to invite employees to a virtual meeting, showing a still image of the executive with no audio or using deepfake audio, claiming technical problems, then instructing transfers through the meeting chat or a follow-up email.

The four steps of a deepfake-assisted BEC attack

Investigators should treat the email and the media as one incident. The sequence is access or spoofing, a changed instruction, apparent corroboration and a transfer. Splitting it into an email incident and a deepfake incident hides the link between the call and the payment.

The best-documented example is the Arup fraud in Hong Kong. A finance employee received a message that appeared to come from the UK-based CFO about a confidential transaction, then joined a video call with what looked like the CFO and other colleagues. The employee made 15 transfers totalling HK$200 million, about US$25 million. Arup confirmed in May 2024 that fake voices and images were used and said none of its internal systems were compromised.

StepWhat the attacker doesRole of deepfakesControl that breaks it
Access or spoofingCompromises or imitates an executive or vendor mailboxNone yet; research of voices and facesPhishing-resistant MFA, mailbox rule monitoring, lookalike domain alerts
Changed instructionRequests an urgent payment or new bank detailsNone or minimalPolicy that email alone never changes payment details
CorroborationSends a voice note, calls, or hosts a short video meetingCloned voice, face swap or still image of the executiveCallback to a number from the vendor master or directory
TransferPushes for release before checks finishRepeated pressure from the fake executiveTwo-person approval, cooling period for new beneficiaries
Where synthetic media appears in a BEC attack and which control interrupts each step.

How big is the business email compromise problem?

Business email compromise is one of the costliest cybercrimes reported to the FBI. The 2024 IC3 annual report recorded 21,442 BEC complaints and US$2,770,151,146 in losses, second only to investment fraud at about US$6.57 billion. Losses were US$2.95 billion in 2023 and US$2.74 billion in 2022, so the problem is steady rather than shrinking.

IC3 does not separate BEC cases that involved deepfakes, so there is no reliable count of how many used synthetic voices or video. FinCEN's November 2024 alert notes reports of criminals using generative AI to impersonate executives or trusted employees and instruct victims to transfer large sums, which shows the technique is in use without measuring its share.

Controls that stop deepfake-assisted BEC

Put the control on the payment change itself. IC3 recommends confirming transfer requests and account changes through a secondary channel or two-factor authentication. Use a phone number from the vendor master file or internal directory, never one printed on the new invoice or supplied in the suspicious thread or meeting invite.

Require two people to approve new beneficiaries and material changes, and apply a cooling period where business allows. Separate the team that edits vendor bank details from the team that releases payments. A familiar executive voice must not be able to waive any of these steps, and staff need explicit permission to delay a payment while they verify, even when the request appears to come from the top.

  • Verify every new or changed payment destination through an independent, pre-registered contact.
  • Treat requests for secrecy or to bypass normal approvals as red flags, regardless of who appears to ask.
  • Alert on unusual timing, amounts, geographies or approver exceptions.
  • Preserve email headers, meeting exports and voice notes for investigation.

What to do in the first hours after a BEC payment

Contact your bank through a trusted number immediately and ask for a recall, then file a complaint at ic3.gov with transfer dates, amounts, account numbers and the receiving institution. The FBI's 2024 report says most Financial Fraud Kill Chain requests initiated by the IC3 Recovery Asset Team involve BEC; speed largely decides whether funds can be frozen.

Then remove the access that enabled the pretext: reset credentials, revoke sessions, delete malicious mailbox rules and review connected apps. Warn affected vendors and staff through verified channels and name the invoices or instructions that are invalid. Do not announce that a recording was synthetic until analysis supports it; the transfer is unauthorised either way.

Where NeuroDetect fits in a BEC investigation

Deepfake detection will not stop a payment that policy allows, but it answers questions the investigation, insurer and law enforcement will ask: was the voice note synthetic, and was the meeting recording manipulated? NeuroDetect, Neural Defend's forensic media workspace, takes authorized audio, video, images and documents into a case and shows results down to audio segments, frames and regions, with an LLM-assisted summary linked to the evidence and a downloadable forensic report that records uncertainty and model version.

For live calls into finance or contact-centre teams, Neural Defend's NeuroCall, currently in Beta, focuses on voice deepfake detection. Treat either result as one input. The payment controls above should hold even if a cloned voice goes undetected.

Evidence boundary

Limitations

  • IC3 figures cover reported complaints only and do not identify how many BEC cases involved synthetic media.
  • Short, compressed voice notes give detectors limited evidence; results should be recorded with their uncertainty.
  • Details of the Arup case come from company statements and press reports; full forensic findings have not been published.

Frequently asked questions

Can a deepfake voice note be used in business email compromise?

Yes. A short voice note in an executive's or vendor's cloned voice is a common way to make a fraudulent email feel confirmed, because it arrives on a different channel. Treat voice notes as unverified and confirm payment changes by calling a number you already hold.

How is deepfake BEC different from ordinary BEC?

The goal and the email are the same: get an unauthorised payment or bank-detail change. The difference is the confirmation step, where a cloned voice, a fake video meeting or a still image of the executive replaces the victim's normal callback. Controls tied to the payment change work against both.

Should finance teams verify executives on video calls?

Yes, when a call asks for money or a change to payment details. Use a pre-agreed verification method or call the executive back on a known number after the meeting, and never release funds based on the call alone.

Can deepfake detection prevent business email compromise?

It can flag synthetic media and support investigations, but it should not be the only barrier. Payment controls such as independent callbacks, dual approval and vendor-master separation stop the fraud even when the media is convincing.

Source record

Sources and references

  1. FBI Internet Crime Complaint Center2024 IC3 Annual ReportPublished April 23, 2025. Verified October 19, 2026.
  2. FBI Internet Crime Complaint CenterBusiness Email Compromise: Virtual Meeting Platforms (I-021622-PSA)Published February 16, 2022. Verified October 19, 2026.
  3. FBI Internet Crime Complaint CenterBusiness Email CompromiseVerified August 7, 2026.
  4. Financial Crimes Enforcement NetworkFinCEN Alert on Fraud Schemes Involving Deepfake Media Targeting Financial Institutions (FIN-2024-Alert004)Published November 13, 2024. Verified October 19, 2026.
  5. Insurance Business AsiaDeepfake video conference sees criminals escape with HK$200 millionPublished May 20, 2024. Verified October 19, 2026.

BEC investigation

Find out whether the voice note was real.

Upload authorized recordings to NeuroDetect for audio-segment and frame-level evidence, with a forensic report for your incident file.

Explore NeuroDetect