Liveness paradox in digital banking exposes onboarding risk

At a glance

What this is: This article argues that digital banking onboarding is being hurt by legacy liveness checks that create friction without reliably stopping modern synthetic fraud and injection attacks.

Why it matters: IAM and identity governance teams need to treat onboarding as a risk-based control design problem, because the wrong biometric flow can suppress conversion while still leaving fraud paths open.

By the numbers:

• 70% of financial institutions globally lost clients in the last year specifically due to slow or inefficient onboarding processes, according to Fenergo research cited by Oz Forensics.
• 38% of customers abandon account creation simply because the process takes too long, according to Deloitte research cited by Oz Forensics.
• Nearly 68% of consumers have abandoned a financial application due to technical friction, according to Oz Forensics.

Context

Digital banking onboarding now behaves like a security control point, not just a user journey. When liveness checks are designed as blanket hurdles rather than risk-based controls, they can reduce fraud exposure in theory while increasing abandonment and operational loss in practice.

The article frames the core issue as governance, not biometrics alone. Financial institutions need to decide when to apply stronger verification, how to keep legitimate customers moving, and which attack patterns demand stream integrity checks rather than more user prompts.

Key questions

Q: What breaks when banks use gesture-heavy liveness as a blanket onboarding control?

A: Gesture-heavy liveness breaks when every applicant is forced through the same friction-heavy path regardless of risk. It can reject legitimate users for poor lighting or device issues while still missing deepfakes and injected video. The control stops being a trust signal and becomes a conversion tax that does not reliably raise assurance.

Q: Why do synthetic media attacks matter more than simple selfie spoofing in digital onboarding?

A: Synthetic media attacks matter because they target the capture pipeline itself, not just the user’s appearance. Deepfakes and virtual camera injection can produce media that looks live enough to defeat gesture prompts. That means the security problem is no longer only identity presentation, but whether the input stream is authentic from the sensor onward.

Q: How should banks balance liveness security with onboarding conversion?

A: Banks should stop treating liveness as a universal challenge-and-response step and instead align verification strength to applicant risk. Use low-friction capture where confidence is high, then escalate only when device, behavioural, or fraud signals justify it. That approach protects conversion without turning identity proofing into an obstacle course.

Q: What is the difference between active liveness and stream integrity checks?

A: Active liveness tests whether a person can complete a challenge such as blinking or turning their head. Stream integrity checks test whether the video came from a real camera and has not been injected or manipulated. Both matter, but they solve different problems, and confusing them leaves a gap in onboarding assurance.

Technical breakdown

Why legacy active liveness creates onboarding friction

Active liveness asks users to perform gestures such as blinking, head turns, or phone movements so the system can test for a live person. That works only when the user experience cost is acceptable and the attack model is simple enough to justify the friction. In banking onboarding, those assumptions break down quickly. If every applicant is treated like a high-risk case, the control becomes a conversion barrier instead of a risk filter. The technical problem is not biometrics in general, but over-reliance on first-generation gesture workflows as a universal proof of presence.

Practical implication: map onboarding steps to risk tiers, not a single fixed liveness path.

How deepfakes and injection attacks bypass gesture checks

Modern synthetic fraud does not depend on the user failing a challenge. Generative AI can power 'puppetry' attacks, where a deepfake avatar mirrors facial responses in real time, and virtual camera software can inject synthetic video directly into the stream. In both cases, the attacker is targeting the integrity of the signal, not the user's ability to complete a gesture. That means a simple smile prompt or blink test no longer defines assurance. The stronger control question is whether the system can distinguish a real camera feed from a synthetic one and detect manipulation before the decision is made.

Practical implication: validate video-stream integrity and synthetic media detection, not gesture completion alone.

Why risk-based orchestration beats one-size-fits-all liveness

Risk-based orchestration treats onboarding as a sequence of controls that can be adjusted to context. Low-friction capture can be used where the risk is lower, while stronger analysis can be reserved for sessions that show signs of manipulation or unusual behaviour. This model is more defensible than choosing active or passive liveness as if they were fixed categories. The article also points to injection attack detection as a separate integrity layer, which matters because liveness and stream provenance solve different problems. One checks whether a face appears live, the other checks whether the data originated from a physical sensor.

Practical implication: separate identity proofing, liveness, and stream provenance into distinct control decisions.

Threat narrative

Attacker objective: The attacker wants to pass onboarding controls with synthetic media and obtain a legitimate-looking banking identity.

1. Entry occurs during digital onboarding when an attacker presents a synthetic face or injects synthetic video into the capture stream.
2. Credentialed or trusted access is not the issue here; the attack instead abuses biometric trust by making a non-genuine session appear legitimate enough to pass verification.
3. Escalation happens when the platform accepts manipulated media as proof of presence and grants account creation or onboarding approval.
4. Impact is fraudulent customer acquisition, polluted identity records, and exposure to downstream financial crime and account abuse.

Breaches seen in the wild

• Internet Archive breach 2024: An exposed GitLab token opened Internet Archive code and 31 million user records; unrotated Zendesk tokens let the attacker back in weeks later.

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

Onboarding liveness has become a governance control, not a UX feature: The article shows that banks are now using biometric assurance as part of account-opening risk management. That changes the decision model from 'does the user complete the check' to 'does the check meaningfully reduce fraud without suppressing legitimate demand'. Practitioners should treat liveness policy as part of onboarding governance, not only identity verification engineering.

Gesture-heavy liveness is an outdated assumption about how fraud presents: The control premise was built for a world where spoofing meant replaying a face or photo. That assumption fails when fraudsters can generate responsive deepfakes or inject synthetic video before the biometric engine sees a real sensor feed. The implication is that teams must rethink where assurance is established in the flow, because challenge prompts alone no longer define trust.

Stream integrity is the named concept that separates assurance from theatre: A system can only trust liveness if it can also trust the provenance of the input stream. The article's injection detection point is important because it shifts the control boundary from user behaviour to capture authenticity. Practitioners should make stream integrity a distinct control objective rather than an implied by-product of face verification.

Risk-based orchestration is the correct design pattern for customer onboarding: The real issue is not active versus passive as a binary choice. Orchestration lets institutions vary control strength by risk, device, and attack signals while keeping legitimate customers moving. That approach aligns better with conversion goals, fraud resistance, and governance accountability than blanket friction ever will.

Financial institutions should measure onboarding controls by both fraud resistance and abandonment: A liveness workflow that stops nothing but frustrates many users is a failed control, even if it looks strict on paper. The article reinforces a dual outcome model where security and conversion both matter. Practitioners should require evidence that onboarding controls reduce attack success without creating disproportionate drop-off.

What this signals

Stream integrity is becoming the real control boundary: Banks can no longer treat liveness as a proxy for capture authenticity. The control that matters is whether the system can distinguish a live camera feed from synthetic or injected video before onboarding approval is granted.

The practical shift for IAM and digital identity teams is to move from user-challenge design to control orchestration. That means separating proofing, liveness, and provenance checks so each can be tuned to a specific fraud condition instead of forcing one step to do all three jobs.

For practitioners

• Segment onboarding by risk tier Use differentiated liveness and verification paths for low-risk, medium-risk, and high-risk onboarding journeys instead of one universal challenge flow.
• Separate gesture checks from stream provenance checks Treat biometric response and camera-feed authenticity as distinct controls so synthetic video injection is evaluated independently from face movement detection.
• Test for deepfake and puppetry abuse Include adversarial testing for responsive synthetic faces, replayed video, and virtual camera injection in onboarding validation exercises.
• Set abandonment thresholds alongside fraud thresholds Track drop-off, false rejects, and fraud-pass rates together so security tuning does not silently erode acquisition.
• Review onboarding controls against account-opening risk Map each identity proofing step to the specific fraud pattern it is meant to stop, then remove steps that add friction without adding assurance.

Key takeaways

• Digital banking onboarding is now a security and conversion decision at the same time, and a single friction-heavy verification path is usually the wrong design.
• Synthetic fraud has moved beyond simple spoofing, which makes gesture prompts insufficient as a stand-alone trust signal.
• The strongest response is to separate risk-based onboarding paths from stream integrity checks so legitimate users are not blocked while manipulated sessions are still caught.

Key terms

• Risk-Based Orchestration: Risk-based orchestration is the practice of adjusting identity verification steps to match the risk level of the session or applicant. Instead of forcing one liveness method on everyone, the flow uses contextual signals to choose the right combination of friction, assurance, and fallback handling.
• 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 Detection: Injection Attack Detection, or IAD, is the ability to identify when a biometric capture stream has been intercepted, replaced, or manipulated before it reaches the verifier. It extends assurance beyond the face or voice itself and tests whether the endpoint, sensor path, and media pipeline can be trusted.
• Server Provenance: Server provenance is the evidence that an MCP server is known, approved, and trustworthy before an agent invokes it. It usually includes registry approval, signed artifacts, and verification of what version is running, which turns server selection into a governed trust decision.

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 building or maturing an IAM programme, it is worth exploring.