Deepfake fraud can bypass identity checks at capture time

At a glance

What this is: NHI Mgmt Group's analysis, drawing on Mitek Systems' blog post: Deepfake fraud is overwhelming identity verification because synthetic media can bypass liveness checks and biometric workflows at capture time.

Why it matters: Identity and fraud teams need layered authentication and authenticity checks because deepfakes target the decision point where a real person is assumed, not the later review stage.

By the numbers:

• Deepfakes were 55% higher than injection attacks in early 2025.
• Mitek Systems says deepfake fraud could increase by 700% more by 2031.
• Mitek Systems cites Deloitte projections that fraud losses from generative AI could rise from $12.3 billion in 2023 to $40 billion by 2027.

Context

Deepfake fraud exploits a trust assumption built into identity verification: that a video selfie, voice sample or live call represents a real person rather than synthetic media. That assumption fails when generative AI can create convincing impersonation at the point of capture, before downstream controls ever see the signal.

This is not just a fraud problem in the abstract. It changes how organisations should think about liveness, biometric verification, document checks and transaction approval because the attacker is no longer only stealing credentials or forging documents; they are manufacturing the identity signal itself.

For financial institutions and other high-trust workflows, the starting position is now atypical: many control stacks were designed for human spoofing and stolen artifacts, not AI-generated media.

Key questions

Q: Why do deepfakes change identity verification requirements?

A: Deepfakes change requirements because they let attackers create believable but false evidence that can pass weak visual checks. Identity verification must therefore move from judging what looks real to verifying whether the entire evidence chain is trustworthy. That includes media authenticity, capture integrity, and stronger assurance criteria for higher-risk transactions.

Q: Why do liveness checks fail against synthetic media?

A: Liveness checks fail against synthetic media because the attacker can present a fake face, voice, or motion signal that matches the challenge well enough to satisfy the verifier. Weaknesses include replay attacks, deepfake video, injected camera streams, and poor challenge design. Stronger controls use active liveness, multi-modal checks, device attestation, and step-up verification against trusted identity evidence.

Q: What should teams do first when deepfake fraud is targeting identity verification?

A: Start with the highest-risk recovery and escalation paths, because those flows often accept weaker proof than initial login. Remove knowledge-based prompts, review call-centre scripts, and require stronger step-up checks wherever an attacker could socially engineer a support agent into resetting access.

Q: What is the difference between presentation attacks and injection attacks in deepfake fraud?

A: Presentation attacks try to fool the system directly with an image, mask, screen replay, or video shown to the camera. Injection attacks bypass the camera path altogether by replacing the live feed with synthetic or prerecorded media. Both aim to impersonate a real person, but the technical control point and detection approach are different.

Technical breakdown

Why capture-stage fraud defeats liveness checks

Liveness checks are designed to answer whether a subject is present and responsive in real time. Deepfakes change the question by attacking the evidence itself, because the submitted video or audio may be synthetic before the system even evaluates motion, gaze or voice patterns. Passive liveness methods look for algorithmic traces such as lighting inconsistencies, eye reflections, texture anomalies and audio sync errors. When those signals are absent or too subtle, a basic control can be satisfied by content that never came from a live person. See the Deepfakes, Social Engineering and AI Impersonation Guide.

Practical implication: Treat liveness as one signal in a broader authenticity stack, not as a stand-alone trust decision.

Why biometric pipelines need layered authenticity checks

Biometric verification works best when it assumes the sample originates from a genuine user at the point of capture. Deepfake fraud breaks that assumption by supplying a synthetic face or voice that can be tuned to pass a single control. That is why layered identity verification combines document checks, passive liveness, behavioural analytics and fraud signals. Each layer raises the cost of reuse and makes evasion more fragile across onboarding, payments and account recovery. Read Top 10 Social Engineering Attacks.

Practical implication: Design identity flows so one successful spoof cannot satisfy every verification layer.

How injection attacks differ from deepfake capture attacks

Injection attacks alter data as it moves through a verification pipeline, while deepfake attacks replace the content at capture with synthetic media. The distinction matters because the same defence does not block both paths. A system can be strong against manipulated transit data and still be vulnerable to AI-generated inputs that are already convincing when they arrive. Governance has to separate the source of the signal from the transport path and tune controls to both.

Practical implication: Map controls separately for capture integrity and pipeline integrity instead of assuming one detector covers both.

Threat narrative

Attacker objective: The attacker wants to persuade a trusted workflow to treat synthetic identity evidence as genuine so money, access or approvals can be diverted.

1. Entry begins with a synthetic video, audio or image created to impersonate a legitimate customer, employee or executive.
2. Credential_harvested activity is replaced by capture-stage deception, where the fraudster submits the deepfake through onboarding or transaction approval flows.
3. Escalation occurs when the synthetic identity convinces staff or systems to override normal verification or payment controls.
4. Impact is financial loss, account compromise or fraudulent transfer, often after the synthetic media is reused across multiple targets.

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

Deepfake fraud is a capture problem before it is a detection problem. The decisive weakness is not that organisations lack more analytics, but that many verification stacks still assume the input is a real human signal. Once synthetic media reaches the capture layer, the control question changes from identity proofing to media authenticity. Practitioners should therefore measure where trust is first established, not where fraud is finally detected.

The real governance gap is overreliance on single-point liveness. A control designed for forged documents or stolen credentials does not automatically handle AI-generated impersonation, especially when the fraudster can tune the content to satisfy one check at a time. This creates a false sense of coverage in onboarding and transaction workflows. Organisations need layered verification because one successful spoof now has outsized reach across the identity journey.

Deepfake fraud creates a reuse economy for attackers. Once a synthetic face or voice is built, it can be iterated and reused across accounts, institutions and approval paths at low marginal cost. That makes the problem structurally different from one-off social engineering attempts and pushes identity teams toward resilient process design rather than reactive model tuning. The named concept here is capture-stage identity compromise: the identity signal itself is manufactured before the verifier sees it, so the control boundary has to move upstream.

Identity verification is converging with fraud operations, not just authentication. The article’s deeper implication is that biometric and document controls now sit inside a broader anti-fraud system that must also account for payment approval, executive impersonation and call-back validation. That expands ownership beyond identity teams alone. Practitioners should align fraud, IAM and operations around the same verification failure modes.

AI-generated media forces a redefinition of what “proof” means in digital identity. If the system only asks whether a face looks real, it will miss whether the media is authentic. That distinction matters for governance, because assurance now depends on source integrity, not just surface realism. Teams should treat authenticity as a distinct control objective in digital identity programmes.

From our research library:

• Businesses report a 200% surge in attempted deepfake-aided wire fraud in Q1 2025 alone.

What this signals

Capture-stage identity compromise: deepfake fraud shifts the defensive problem upstream, because the system may be asked to trust synthetic evidence before any downstream control can intervene. That means verification architecture has to evaluate source authenticity as a first-class risk, not as a cosmetic enhancement to existing checks.

Deepfake resilience will increasingly depend on layered decisioning, not one-pass scoring. Organisations that separate media authenticity, behavioural confidence and transaction approval will be better placed to absorb AI-generated impersonation without turning every verification flow into a brittle gate.

For practitioners

• Implement layered identity verification Combine passive liveness, document verification, behavioural analytics and biometric checks so one evasion path cannot satisfy the entire process.
• Add out-of-band approval for high-risk transactions Require a separate callback or second channel before wire transfers, account changes or executive-driven payment requests are approved.
• Test verification flows against synthetic media Exercise onboarding and payment journeys with deepfake video, audio and image scenarios to find which controls fail at capture time.
• Separate capture integrity from transport integrity Review whether controls detect manipulation at the point of capture, during transmission, or both, then tune monitoring to cover each stage explicitly.
• Train approvers to challenge identity-based urgency Use executive impersonation scenarios to reinforce that urgency, familiarity and visual realism are not sufficient evidence for approval.

Key takeaways

• Deepfake fraud is now exploiting the assumption that captured media is real, which makes traditional identity verification controls insufficient on their own.
• The article ties rapid growth to both easier generation tools and rising financial impact, with deepfake-driven losses projected to climb sharply by 2027.
• Layered verification, out-of-band approval and synthetic-media testing are the practical controls most likely to reduce exposure in high-trust workflows.

Related NHI Mgmt Group resources: Deepfakes, Social Engineering and AI Impersonation Guide · Top 10 Social Engineering Attacks · Deepfake CEO Fraud Response Playbook · Executive Deepfake and CEO Fraud Guide

Key terms

• Deep Fake Fraud: Deep fake fraud is deception that uses synthetic audio, video, or images to impersonate a real person or create a convincing false event. It relies on manipulated media plus social engineering, often to bypass human verification, authorise payments, or spoof identity checks where weak challenge steps are used.
• 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.
• Passive Liveness: Passive liveness is a verification method that checks whether a biometric sample appears to come from a live human without requiring the user to perform actions. It reduces friction, but it still depends on the integrity of the capture and analysis path. If that path is manipulated, passive liveness alone will not protect the decision.
• Biometric Verification: Biometric verification confirms a person’s identity by matching live physical or behavioral traits, such as a face, fingerprint, voice, or iris, against a previously enrolled reference. In IAM, it is an authentication method that can support step-up access, but requires strong liveness checks, spoof resistance, and privacy controls.

Deepen your knowledge

The NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, machine identity security and secrets management. It helps security practitioners connect identity controls to the broader governance decisions their programmes depend on.