Skip to content
Authentication & identity · Financial services

Voice AI authentication and identity in financial services

Authentication is the hardest half of every banking voice AI call. The deployments that survive a fraud post-mortem treat the calling number as untrusted, tier assurance by the action requested, and use voice biometrics as a factor inside SCA — never as a substitute for it.

Realistic containment band

Not a containment use case — measure success, friction, fraud loss

Integration touchpoints

  • Identity service exposing tiered assurance decisions on a per-action basis
  • Voice biometrics platform integrated as an inherence factor, with enrolment governance
  • Push-to-mobile / OTP channel for active step-up that does not depend on the SIM
  • Fraud platform write-through for AI-observed risk signals (cadence, ANI mismatch, behavioural)

Regulatory hooks

  • PSD2 SCA — independent factors from at least two of knowledge / possession / inherence
  • FCA Consumer Duty (UK) — false-reject rates that cluster on demographic must be measured and remediated
  • GDPR Article 9 — biometric data is special-category; explicit consent and DPIA required
  • UK Contingent Reimbursement Model on APP fraud — authentication strength is a defence and a liability driver

What good looks like

Tiered assurance encoded as policy, not as prompts. Disclosure runs at low; account changes at medium with a passive factor; SCA-grade actions at high with an active factor that does not depend on the SIM. Step-up rate, false-accept, and false-reject are published operational metrics with demographic breakdowns. Voice biometrics is enrolled with informed consent and re-validated periodically.

Watch-outs

  • Trusting the calling number for any change-of-contact action. SIM-swap and number-porting fraud assume you do.
  • Enrolling voice biometrics without explicit informed consent — GDPR Article 9 is unambiguous.
  • Identical assurance for disclosure and change. A balance read and a beneficiary add are not the same risk.
  • Hiding step-up rate from the operations dashboard. A rising step-up rate is an early signal of an attack pattern.

Frequently asked

Does voice biometrics satisfy PSD2 SCA on its own?

No. Voice biometrics qualifies as the inherence factor inside SCA, but SCA requires two independent factors. The deployments that hold up pair voice biometrics with a knowledge or possession factor on every SCA-grade action.

How do you design for SIM-swap fraud?

Treat the SIM as untrusted by default. Any sensitive change requires a factor that does not depend on the SIM — push-to-app to a previously-enrolled device, OTP to a verified second channel, or in-branch. The AI's job is to enforce that policy, not to negotiate it.

Related