Facial deepfakes are outpacing biometric identity verification

At a glance

What this is: This is Incode's breakdown of three common facial deepfake methods and its core finding that human visual review is no longer a dependable control for biometric identity verification.

Why it matters: It matters because IAM, KYC, and fraud teams that still depend on human inspection or shallow liveness checks are exposed to scalable impersonation attempts that can bypass identity assurance.

Context

Facial deepfakes are now a biometric identity problem, not just a content authenticity problem. Incode's article explains how synthetic faces, face swaps, and face animation are being used to bypass identity verification and biometric authentication, especially where the face is treated as the primary identifier.

The governance gap is straightforward: controls designed around human inspection and obvious visual artifacts no longer match the quality of current AI-generated media. For teams running KYC, identity proofing, or customer onboarding, the question is no longer whether a deepfake can look real, but whether the verification flow can still distinguish a live person from a generated session.

Key questions

Q: What breaks when biometric verification still depends on human visual inspection?

A: Human review breaks because current deepfakes can look realistic enough to defeat artifact hunting, while low-quality real images can trigger false suspicion. The result is both missed fraud and unnecessary rejection of legitimate users. Verification needs machine-backed signals, not eyeballing alone.

Q: How can organisations reduce fraud from synthetic faces and face swaps?

A: Shift from standalone verification decisions to session-level risk evaluation. Correlate capture quality, camera behaviour, document checks, and repeat patterns across attempts so synthetic identities and swapped faces are easier to separate from legitimate users.

Q: How can identity teams tell whether liveness checks are actually working?

A: Look for adversarial cases that still pass when motion is synthetic, partial, or replayed from a static image. Effective programmes test whether liveness outcomes remain stable when the same applicant reappears with different devices, names, or document data. If the control only succeeds in clean test conditions, it is not robust against fraud reuse.

Q: What should organisations do when face animation can mimic a live selfie session?

A: They should treat motion as a signal, not proof, and require the decision engine to combine liveness with device telemetry, session integrity, and fraud risk signals. That reduces the chance that an animated image is accepted as a genuine person.

Technical breakdown

Synthetic faces and why generation scale changes fraud economics

Synthetic faces are AI-generated identities that do not need to correspond to a real person. Modern diffusion and GAN-based models can produce realistic still images and increasingly convincing video, which lets attackers create large volumes of fake identities cheaply and quickly. That matters because the attack is not limited to better realism. It also changes the economics of fraud, enabling bulk registration attempts and repeated verification runs until a path succeeds. In biometric workflows, this means the problem is not merely artifact detection. It is adversarial media generation at scale, combined with identity creation pipelines that can be automated end to end.

Practical implication: treat synthetic media as a scale problem and validate onboarding flows against mass-generated identity attempts, not just single-image spoofing.

Face swaps and the limits of visual inspection

A face swap overlays one identity onto another person’s image or video, often using a target recording that already contains believable lighting, pose, and motion cues. Earlier methods left visible artifacts, but current generative tooling can preserve enough realism to evade casual review. Incode's point is that human reviewers can no longer depend on obvious mismatches such as edge artifacts or unnatural backgrounds. When the swapped face is paired with a stolen or fake ID document, the workflow becomes a layered impersonation attempt rather than a simple spoof. The control failure is the assumption that trained eyes can reliably arbitrate authenticity in every session.

Practical implication: move from reviewer judgment alone to multi-signal verification that correlates document, device, and liveness evidence in the same session.

Face animation and digital injection into verification pipelines

Face animation, also called face reenactment, turns a still image into a moving selfie-like video by adding blink, motion, speech, or expression cues. This is especially relevant to identity verification because the output can mimic a legitimate capture session, even when the source image is synthetic or stolen. Incode also notes that some attacks are injected into verification pipelines rather than captured live, which means the system must defend the session itself, not just the media content. The deeper issue is that the presence of motion no longer proves presence. For identity assurance, motion is just another attribute that can be synthesized.

Practical implication: assess whether your liveness control can distinguish live capture from injected media, not only static spoofing from real faces.

Threat narrative

Attacker objective: The objective is to bypass biometric verification and establish fraudulent trust under a fake or stolen identity.

1. Attackers generate or obtain synthetic faces, face swaps, or animated selfie videos and use them as initial presentation material for identity verification flows.
2. They combine the deepfake with fake or stolen identity artifacts and submit it through onboarding or authentication pipelines that still rely on visual trust cues.
3. The verification process accepts the session as legitimate when human review or lightweight liveness checks fail to distinguish generated media from a real capture.
4. Successful impersonation enables account creation, onboarding fraud, or access under a false identity at scale.

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

Human visual review is no longer a security control for biometric identity. Deepfakes have crossed the threshold where even trained labelers cannot consistently separate synthetic media from real capture. That means review remains useful as a backstop, but not as a primary authentication decision point. The practitioner conclusion is simple: if the process depends on a human spotting artifacts, the control is already behind the threat.

Biometric verification now depends on session trust, not image trust. The article shows that fraudsters are not only generating faces, they are injecting media into verification pipelines. That shifts the governance problem from static authenticity to runtime assurance, where device signals, capture context, and liveness evidence matter more than what the face looks like. The practitioner conclusion is that identity assurance has to be built around the session boundary.

Deepfake-driven identity fraud is a lifecycle problem, not a point-in-time detection problem. Synthetic faces can be created, reused, and iterated until one passes, which means issuance and enrolment controls matter as much as detection. This is where KYC, identity proofing, and fraud operations intersect. The practitioner conclusion is that programs need to govern the entire identity event, not only the final verification frame.

Face animation collapses the assumption that motion implies presence. Once blinking, speech, and expression can be generated from a still image, the common heuristic that live movement equals live capture stops holding. That is a control-premise failure, not just a model-quality issue. The practitioner conclusion is to rethink any workflow that treats movement alone as sufficient evidence of a real person.

From our research library:

• Veriff's 2026 Identity Fraud Report documented a 300X increase in digitally presented media that was either entirely AI-generated or otherwise altered.

What this signals

Session trust is now the control boundary. When deepfakes can be injected into a verification flow, the security question shifts from whether a face looks real to whether the session itself was created, transported, and captured legitimately. That is a stronger framing for identity proofing teams than trying to out-hunt every new artifact.

Identity proofing and fraud prevention have to converge. The article shows that face generation, face swap, and animation are different technical paths to the same outcome: false enrolment or impersonation. Programs that separate onboarding assurance from fraud controls leave a gap that attackers can exploit with repeated attempts.

Deepfake pressure makes liveness a composite decision. A robust workflow will increasingly combine image analysis, device intelligence, and behavioral context rather than treating any one signal as authoritative. For practitioners, that means the operating model changes as much as the detection stack.

For practitioners

• Harden biometric onboarding against generated media Test onboarding and recovery flows with synthetic faces, face swaps, and face animation so the control stack is evaluated against current attack quality, not older spoofing methods.
• Correlate liveness with device and session signals Require the verification decision to use device posture, capture context, and behavioral signals alongside liveness so the system is not relying on one media channel alone.
• Reduce trust in visual review alone Recalibrate reviewer playbooks so human analysts are used for escalation and exception handling, not as the primary authenticity gate for biometric identity.
• Validate protections against injected verification media Simulate pipeline injection scenarios, including prerecorded selfie videos and animated still images, to confirm the verification stack can distinguish live capture from replayed or manipulated media.

Key takeaways

• Facial deepfakes are no longer easy-to-spot visual tricks. They now support scalable identity fraud that can defeat shallow biometric controls and human review.
• The article's core evidence is operational, not theoretical. Incode says even experienced labelers cannot consistently identify deepfakes by sight alone, which means confidence in visual inspection is misplaced.
• Practitioners need verification that combines liveness, device, and session signals. If the control only checks what the face looks like, it is already behind the threat.

Key terms

• Face swap: A face swap replaces the face in a real image or video with another identity while preserving the surrounding scene and action. The control challenge is that the body, lighting, and background can remain authentic-looking even when the identity layer has been substituted.
• Synthetic Face: A synthetic face is an AI-generated face that does not belong to a real person but is designed to look authentic. For biometric security, it matters because it can be produced at scale and used to create false identities that have no real-world owner to verify.
• Face animation: Face animation adds motion, blinking, speech, or expression to a still image so it appears live and interactive. For identity teams, the risk is that a static or stolen image can be turned into convincing capture media that defeats weak liveness checks.
• 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.

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.