Why do deepfakes create fraud risk even when MFA is already deployed?

Why MFA does not stop deepfake fraud in a live conversation

MFA is designed to raise the cost of account login, not to verify the person speaking after the session begins. In a voice call, video meeting, or message thread, the attacker can exploit trust directly, using a convincing synthetic persona to request urgency, secrecy, or a payment action. That makes the control gap one of communication-time identity, not only sign-in identity.

The practical problem is that many fraud workflows happen outside the application boundary. Once the attacker has access to the right channel, the victim often relies on recognition, tone, job title, or apparent familiarity rather than a second cryptographic proof. Deepfakes are effective precisely because they imitate the social evidence humans use to grant confidence.

Where the fraud path opens after login succeeds

Deepfakes matter because they can be paired with other access methods and then used to turn routine business communication into an approval channel. A fraudster may not need to defeat MFA if they can persuade someone to disclose information, alter bank details, approve an invoice, or bypass a normal verification step. The weakness is not the login alone, it is the assumption that authenticated access implies trustworthy intent.

This is why payment, payroll, treasury, and help-desk style workflows are common targets. The attacker only needs one person to treat an identity claim as sufficient evidence. When that happens, MFA has done its job at the door, but the organisation has no equivalent control at the moment of decision.

What organisations need in the second assurance layer

The answer is to separate sign-in assurance from communication assurance. Use out-of-band verification for material requests, require callback or codeword checks for sensitive changes, and add policy friction where a single conversation can trigger financial or privileged action. For broad workforce identity controls, NHIMG’s Workforce Identity Security Guide is useful because it connects phishing-resistant sign-in with the recovery and session risks that remain after authentication.

Where the concern is specifically impersonation and fraud during calls or video, the right control set is closer to human verification than access management. NHIMG’s Deepfakes, Social Engineering and AI Impersonation Guide and Arup deepfake fraud 2024 both show why a live voice or video presence cannot be treated as proof of legitimacy. The control objective is to force confirmation through an independent channel before value moves.

Risk and Threat Considerations

Deepfakes create fraud risk because they let an attacker borrow authority from a familiar face or voice while avoiding the friction of a normal login attack. The danger is highest where staff are trained to respond quickly, defer to senior people, or approve exceptions under time pressure.

Failure mechanism: The attacker exploits human trust in real-time conversation, then converts that trust into a financial transfer, credential reset, data disclosure, or policy exception. MFA does not interrupt that path once the target is already engaged in the call or meeting.

Impact: Organisations can suffer unauthorised payments, business email compromise style losses, account recovery abuse, and downstream access expansion if the victim helps the attacker reset or rebind controls.

Practitioner Guidance

What to prioritise: Treat any request that changes money movement, beneficiary details, login recovery, or privileged access as a high-risk event, even when the requester sounds familiar. The control should be strongest at the point of action, not only at sign-in.

What to verify: Require a verification step that is independent of the current call, meeting, or message thread. If a request can be executed with only conversational trust, it is under-controlled.

Decision rule: If the action would create irreversible loss or expose a privileged path, stop and re-verify through a known-good channel before proceeding.

Practitioner takeaway: MFA reduces one class of impersonation, but deepfake fraud succeeds when organisations fail to authenticate the conversation itself.

Related resources from NHI Mgmt Group

• Why does security product drift create risk even when tools are already deployed?
• Why do non-human identities create more risk than many human accounts?
• Why do non-human identities create more remediation risk than many human accounts?
• Why do machine identities create risk even when MFA is enabled?