Deepfakes are testing identity controls beyond visual trust

At a glance

What this is: This article explains how increasingly realistic deepfakes are being used for employee impersonation, account takeover, and synthetic identity fraud, and why visual trust is no longer enough.

Why it matters: Identity, IAM, and fraud teams need to treat deepfakes as an authentication problem, because human judgment and face or voice checks can be bypassed under live attack conditions.

Context

Deepfakes create a trust problem for identity programmes because the signal a human reviewer relies on, what they can see or hear, is no longer reliable on its own. The article is about synthetic media used to impersonate people during calls, interviews, and account verification, which pushes the risk directly into identity assurance and fraud controls.

For IAM and fraud teams, the issue is not only whether a clip looks fake after the fact. It is whether identity verification, MFA, liveness detection, and participant verification can still establish trust while the attacker is actively shaping the interaction in real time.

Key questions

Q: What breaks when identity verification depends on the user spotting fraud?

A: The control fails when the person being protected becomes the detection layer. Human review is slow, inconsistent, and easy to bypass when attacks look legitimate and arrive continuously. That creates a gap between the speed of the fraud and the speed of the response, which is why silent verification is gaining value.

Q: Why do deepfakes make traditional authentication weaker?

A: Deepfakes weaken traditional authentication because they imitate the human signals that many approval processes still trust, including voice and video. When those cues can be fabricated, organisations need independent verification paths such as out-of-band confirmation, device checks, and transaction-specific controls for high-risk actions.

Q: How do security teams evaluate whether liveness detection is strong enough?

A: Look for measurable resistance to presentation attacks, defined false accept and false reject rates, and testing that reflects the real environment where the control will run. A good evaluation also includes exception handling, logging, and whether staff can bypass the control when operations become urgent.

Q: How should organisations adapt fraud controls as deepfake attacks become more convincing and more common?

A: Organisations should treat deepfake fraud as a multi stage risk, not just an onboarding problem. Effective controls combine liveness checks, document verification, behavioural signals, step up review, and ongoing monitoring through the customer lifecycle. Teams also need clear escalation paths for high risk cases, because synthetic media can defeat single point checks and create false confidence in identity assurance.

Technical breakdown

Why deepfakes now defeat visual trust

Deepfakes are synthetic audio, video, or images generated by machine learning models that can mimic human appearance and speech with increasing fidelity. The article highlights the shift from obvious artefacts to near-photorealistic output, driven by deeper training data, better generative models, and real-time synthesis. That matters because identity decisions often depend on a person appearing credible during a live interaction, not on cryptographic proof. Once the media stream itself can be manufactured on demand, the control fails before the human reviewer knows they are being deceived.

Practical implication: treat visible likeness as an untrusted signal and require stronger verification than human observation alone.

How injection attacks bypass camera-based checks

Injection attacks target the verification channel rather than the person. Instead of presenting a face to the camera, an attacker inserts pre-recorded or synthetic media directly into the authentication flow, bypassing the live sensor input that liveness controls expect to inspect. That creates a different failure mode from simple spoofing, because the system may believe it is analysing a real camera feed when it is actually receiving manipulated content. In identity terms, the control is not only checking the wrong subject, it is checking a channel that has already been subverted.

Practical implication: verify that the capture path itself is protected, not just the biometric outcome it produces.

Why liveness and MFA need to work together

Liveness detection checks that a real person is present, while MFA adds a separate factor that should not depend on the same compromised media channel. The article’s core point is that face matching alone is insufficient when deepfakes can imitate a user in real time, especially during remote onboarding or account recovery. Identity assurance becomes stronger when controls are layered so that one control addresses impersonation and another addresses session or account access. That is the practical boundary deepfakes expose: single-signal verification collapses when fraud can manufacture the signal.

Practical implication: combine liveness checks with an independent second factor so impersonation cannot ride the same channel twice.

Breaches seen in the wild

• Arup deepfake fraud 2024: Deepfakes of Arup's CFO and colleagues on a video call led a Hong Kong employee to transfer HK$200 million (about US$25.6m) to fraudsters.

Read and download The State of NHI & AI Agent Breach Report 2026, covering 200+ breaches impacting Non-Human Identities including AI Agents.

NHI Mgmt Group analysis

Visual trust is no longer a control boundary. The article shows that identity programmes built on “if it looks like the user, it is the user” have crossed a practical limit. Deepfakes now let an attacker manufacture the human signal that many verification flows were designed to trust. The implication is that identity assurance must shift from appearance-based judgment to stronger proof of presence and control.

Real-time impersonation collapses the review window. Deepfake fraud is not delayed or static. It can unfold inside a live call, interview, or verification flow before a human reviewer has time to challenge it, which means the control must operate during the interaction, not after it. For practitioners, this is a governance problem as much as a detection problem because the attack is designed to outrun manual review.

Participant verification is becoming a first-class identity control. The article’s business examples show that identity checks are now protecting meetings, onboarding, access recovery, and high-risk approvals, not just user login screens. That broadens the scope of IAM and fraud teams, because the question is no longer only who can sign in, but who can plausibly participate in a trusted interaction. Organisations should treat participant verification as part of identity governance, not a niche fraud add-on.

Deepfake detection is now an anti-spoofing capability, not a branding feature. The presence of real-time generation and injection attacks means defenders need controls that validate the integrity of the capture path, the liveness of the subject, and the independence of the second factor. This is where face match alone fails, because it answers the wrong question. The practical conclusion is that identity programmes need fraud-resistant verification workflows, not just more polished user experiences.

From our research library:

• According to a recent Gartner report, 40% of government organizations are expected to establish dedicated trust operations functions by 2028 to combat deepfakes.

What this signals

Identity proofing is moving closer to fraud operations. Deepfake pressure means IAM teams can no longer treat verification as a one-time front door event. The programme has to assume that an attacker may be present during the interaction itself, which makes capture integrity, liveness, and step-up controls part of the same decision chain.

Participant verification will become a standard control in high-trust workflows. As synthetic media improves, organisations will need stronger checks around meetings, onboarding, recovery, and payment approvals. That shift matters because the weak point is no longer just the login page, it is any workflow that depends on a trusted human being visibly present.

For practitioners

• Harden verification beyond visual checks Require additional proof before accepting a video, voice, or image as evidence of identity. Use trusted identity profiles, stronger enrolment checks, and layered verification so a convincing synthetic face or voice cannot complete the process on its own.
• Add liveness checks to high-risk flows Use liveness detection for onboarding, account recovery, and any interaction where a remote participant can influence access or payment decisions. Make the control independent from simple face matching so the system verifies a live person, not only a realistic image.
• Protect the capture path from injection Validate whether a camera feed has been overridden or substituted before it reaches the decision engine. Defend both desktop and mobile capture paths so pre-prepared media cannot be injected into the authentication flow.
• Separate impersonation resistance from access control Treat deepfake resistance as one control layer and MFA as another. If one layer is bypassed, the second factor still has to block access to systems, accounts, or approvals that would otherwise be exposed.

Key takeaways

• Deepfakes have turned appearance and voice into weak trust signals for identity and fraud teams.
• The highest-risk failure point is live interaction, where synthetic media can defeat review before anyone has time to react.
• Controls that combine liveness, MFA, and capture-path protection reduce the chance that one convincing impersonation can reach access or approval.

Key terms

• Deepfake: Synthetic or altered media created with AI or machine learning so that a person appears to say or do something they never did. In security terms, deepfakes are trust attacks that can distort identity verification, approval workflows, and fraud detection.
• Liveness Detection: Liveness detection is the mechanism that checks whether a biometric sample comes from a real, present person rather than a spoof such as a photo, screen, or mask. In identity programmes, it is a core defence against presentation attacks and should be tested under realistic operating conditions.
• Injection attack: An attack that inserts synthetic or manipulated data directly into the verification flow rather than fooling the sensor itself. For identity programmes, this is a control-path problem, because the attacker may bypass the visible presentation layer and exploit the software decision point.
• Participant Verification: Participant verification is the process of confirming that each person in a meeting, call, onboarding session, or approval workflow is genuinely who they claim to be. It extends identity assurance beyond login because fraud increasingly happens during live interaction, not only at sign-in.

Deepen your knowledge

NHI governance, agentic AI identity, and machine identity lifecycle are core topics in our NHI Foundation Level course, the industry's only accredited NHI security programme. If you are responsible for identity security strategy or NHI governance in your organisation, it is worth exploring.