Biometric defense against app fraud liability shift in banking

At a glance

What this is: This is an analysis of how APP fraud liability shifts are changing the control expectations on banks, with biometric liveness and injection-attack detection positioned as the practical response to deepfake-enabled onboarding and payment fraud.

Why it matters: It matters because IAM and fraud teams now have to treat identity verification as a loss-prevention control, not just an onboarding step, and prove that biometric signals are live, not replayed or injected.

By the numbers:

• UK banks reimbursed £112 million to victims, with a reimbursement rate hitting 88%.
• The UK introduced a 50/50 liability split for fintechs between the sending bank and the receiving bank.
• Oz Forensics says solutions certified against CEN/TS 18099 for Injection Attack Detection are the 2026 essential.
• The company claims its lab results reached 100% detection accuracy in BixeLab.

Context

APP fraud liability is changing the economics of digital banking. Mandatory reimbursement rules mean that banks can no longer treat authorised payment scams as a customer-only problem, because the institution may now carry the loss even when the customer authorised the transfer under deception.

That shift raises the bar for identity assurance at onboarding and payment approval. Legacy face checks are not enough when attackers can use deepfakes, emulators, virtual camera software, and injected video streams to create mule accounts or drive fraudulent approvals through mobile banking apps.

Key questions

Q: What breaks when banks rely on face matching without liveness detection?

A: Face matching alone breaks when an attacker can inject a replayed, synthetic, or manipulated video stream into the app. The control may still see a plausible face, but it has no trustworthy evidence that the person is live or that the capture path was untampered. That creates a false sense of assurance in onboarding and payment approval.

Q: Why do APP fraud reimbursement rules change identity control priorities?

A: They convert identity failure into a direct financial liability. That means onboarding verification, liveness testing, and payment approval controls should be evaluated by how well they prevent reimbursable losses, not only by how many fraud cases they detect later.

Q: How can banks tell whether liveness controls are actually working?

A: They should test whether the control rejects emulators, virtual camera feeds, replayed sessions, and injected synthetic video under realistic fraud tooling. A useful liveness control must fail closed when the capture path is manipulated, not just when the face is poor quality. Lab validation and red-team testing are stronger signals than vendor claims alone.

Q: Should banks compare passive liveness with active challenge flows?

A: Yes, but the comparison should be based on fraud risk, user friction, and the strength of the capture-path assurance required. Passive liveness reduces abandonment and still raises the bar against spoofing, while active challenge flows can add friction but may help in higher-risk journeys. The right choice depends on the liability exposure of the flow.

Technical breakdown

Why passive liveness matters in banking onboarding

Passive liveness checks try to prove that a biometric sample comes from a real person in real time without adding a manual challenge. The article describes analysis of micro-reflections on skin, depth cues, and indicators that a camera feed is a screen, mask, or photo. In banking, that matters because the attacker is often not trying to spoof every control, only the one that turns an account into a usable mule account. Passive liveness reduces friction while still raising the cost of synthetic enrolment.

Practical implication: treat liveness as an enrolment control that must be resilient to replay and camera injection, not as a user-experience feature.

How injection attacks bypass biometric controls

An injection attack does not rely on presenting a live face to the camera. Instead, the attacker uses emulators, virtual camera software, or a manipulated app stream to feed pre-recorded or synthetic video into the verification process. That defeats systems that evaluate only the visual face image, because the application is being shown a trusted-looking signal rather than observing a live capture path. The technical failure is not facial recognition itself, but the trust boundary around the capture channel.

Practical implication: verify the integrity of the capture path and not just the biometric artefact, or the app will accept synthetic inputs as genuine.

Why certified injection attack detection changes the control baseline

The article frames CEN/TS 18099 as a certification target for injection attack detection, which shifts the discussion from generic biometric matching to detection of manipulated input channels. That is important because the fraud pattern is industrialised, with deepfakes, device farms, and emulator tooling used at scale. In practice, the control objective becomes proving that the biometric signal originated from a live, untampered interaction. Certification and lab validation matter because they create a measurable threshold for a control that would otherwise be hard to benchmark.

Practical implication: require evidence that biometric controls detect injected sessions, not just that they match faces accurately.

Threat narrative

Attacker objective: The objective is to create bank-accepted identities and fraudulent payment approvals that convert stolen access into reimbursable financial loss.

1. Entry begins when fraudsters use synthetic identities, device farms, emulators, or deepfake-assisted social engineering to reach the banking app or onboarding flow.
2. Credential or signal access follows when the attacker injects pre-recorded or manipulated video into the verification stream, bypassing checks that only inspect the presented face.
3. Escalation occurs when the fraudulent identity is accepted as genuine, allowing mule account creation or payment approval that the bank will later have to reimburse.
4. Impact is the successful laundering of stolen funds or the execution of authorised push payment fraud at a scale that shifts losses onto the institution.

Breaches seen in the wild

• Zacks breach claim 2025: A hacker leaked 12 million Zacks accounts in 2025, claiming domain admin access in 2024; HIBP verified the data, Zacks has not confirmed.

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

Liability shift changes the identity control problem, not just the fraud budget. APP fraud reimbursement rules move banks from loss avoidance to loss containment under proof pressure. That changes identity verification from a front-end onboarding check into a financial control that must withstand adversarial review. Practitioners should judge biometric programmes by whether they can support liability defence, not only conversion.

Fake-face detection is now too narrow a control model. The article's core warning is that attackers do not need to spoof a face if they can spoof the session. That means the real governance gap is not image quality, but trust in the capture channel and the device path. Banks that focus only on match scores will miss the higher-value control failure: accepting manipulated input as live identity evidence.

Certified liveness becomes a reimbursement control, not a biometric feature. CEN/TS 18099 matters because it reframes the control from identity convenience to evidential resilience. Once reimbursement frameworks make banks accountable, they need controls that can prove the interaction was live, untampered, and session-authenticated. The practitioner conclusion is simple: if the control cannot stand up in a liability dispute, it is not mature enough for APP fraud exposure.

Identity assurance and fraud operations are converging around the same signal. Banks can no longer separate onboarding integrity from payment-authorisation risk because the same synthetic identity can serve both mule-account creation and APP fraud approval. That convergence means fraud teams and IAM teams need a shared model for live capture, replay resistance, and device integrity. The implication is a single trust policy for the full account lifecycle, not separate controls for enrolment and transfer.

Biometric liveness is becoming a governance boundary for fintech liability. The 50/50 liability split described in the article makes receiving institutions financially responsible for what used to be somebody else’s fraud. That pushes governance upstream into identity proofing, because every weak enrolment path now carries direct loss exposure. Practitioners should treat liveness assurance as part of financial accountability, not just digital onboarding.

What this signals

APP fraud liability is now an identity governance problem. The bank’s obligation is no longer limited to detecting suspicious payments after the fact. It has to govern the evidence used to create and authorise those payments, which makes capture integrity, liveness validation, and fraud operations part of the same control plane.

Capture-path trust debt is the new control gap. If the organisation cannot prove that biometric input came from a live device interaction, it has a governance gap even when the face match succeeds. That gap will matter most in high-volume onboarding and payment journeys where synthetic identities can be converted into reimbursable loss within minutes.

For practitioners

• Harden the biometric capture path Require controls that detect emulators, virtual cameras, and replayed video before the biometric sample reaches match logic.
• Validate liveness against injection abuse Test onboarding and payment journeys with deepfakes, injected streams, and synthetic identities so that failure modes are visible before attackers scale them.
• Tie fraud loss exposure to verification assurance Align reimbursement risk owners with the teams that govern biometric evidence, so control gaps are evaluated as financial liabilities rather than UX defects.
• Define a lab-backed assurance threshold Use certification evidence such as injection-attack detection results to decide whether a control is strong enough for high-liability banking flows.

Key takeaways

• APP fraud reimbursement shifts mean that banks now own more of the loss, so biometric assurance has become a financial control as much as an identity control.
• Deepfakes and injection attacks are effective because they attack the trust boundary around the capture path, not just the face image itself.
• Banks should test liveness controls against synthetic identity tooling, emulator-based injection, and replay abuse before treating them as ready for liability-heavy flows.

Key terms

• 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.
• 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.
• APP Fraud: Authorised push payment fraud happens when a victim is manipulated into approving a payment to a criminal account. Because the payment is user-authorised in form, reimbursement and accountability rules often shift the burden onto banks, making identity assurance at the point of approval financially critical.
• Mule Account: A mule account is an identity or account used to receive, move, or obscure illicit funds on behalf of another party. In financial crime operations, it is the bridge between the initial deception and the laundering phase, often appearing legitimate until behavior reveals coordination.

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.