DeepBlocker← Back
Industries · Financial Services

Voice fraud defence for banks, fintechs and treasury operations.

Voice cloning takes one minute of audio. The result: helpdesk impersonation, CFO fraud, password-reset abuse and synthetic-identity onboarding at carrier-grade scale. DeepBlocker is the only voice deepfake detector with a published model card that holds up on real phone lines.

Book a fraud briefingReal-Time Protection datasheetDetect API datasheet
The threat picture

Voice is now the primary channel for high-value financial fraud.

  • £1.17B stolen from UK victims to authorised push-payment and impersonation fraud (UK Finance, 2025).
  • $16.6B in reported losses to cybercrime, with vishing and impersonation accounting for the fastest-growing categories (FBI IC3, 2024).
  • 60% of firms experienced at least one deepfake-driven incident in the last twelve months (Thales, 2026).
  • Voice authentication is "fully defeated", per Sam Altman addressing the US Federal Reserve in 2024. Most banks still rely on it.
What we defend

The three voice-attack surfaces inside a financial institution.

Helpdesk and customer-service lines

Account takeover via password resets, address changes, and SIM-swap requests. DeepBlocker intercepts the call, verifies the speaker, and pushes high-risk requests to verified processes — never approves a sensitive change by phone.

Real-Time Protection · per-line
Treasury, payments and CFO fraud

Spoofed CEO/CFO calls instructing wire transfers or vendor changes. DeepBlocker terminates synthetic-voice calls before they reach the human, and routes the audit trail to your fraud team.

Real-Time Protection · executive lines
Onboarding and identity verification

Synthetic-voice fraud during phone-based KYC and re-verification. DeepBlocker Detect plugs into your existing fraud-decisioning stack as a hosted API — same telephony-tuned model that powers Real-Time Protection.

DeepBlocker Detect · API
The detection-grade you can audit

Reproducible, telephony-tuned, model-card-grade.

98.5%
ElevenLabs deepfake mean confidence
94.7%
Deepfake recall on real telephony
0.9%
Review-queue load
≥ 0.92
ROC AUC, locked Reporting set

Measured on a held-out Reporting set, n ≈ 634, SHA-256 pinned, used exactly once. We publish methodology and per-source breakdown — including known limitations — in the model card. Onecom-grade due-diligence is reproducible from the open base model.

Regulatory alignment

Built for FCA, DORA, NIS2, and GDPR obligations.

  • FCA Consumer Duty: evidence of meaningful steps to prevent foreseeable harm — every blocked deepfake call has a complete audit trail.
  • DORA (EU 2554/2022): ICT third-party risk management and operational resilience. DeepBlocker contracts include sub-processor disclosure and exit-plan support.
  • NIS2: incident reporting hooks into existing SOC tooling. Per-call disposition logs are exportable to Splunk / Sentinel / Elastic.
  • GDPR: per-tenant retention, encryption, and access controls; data-residency options on request.
The Enterprise Assessment

A 2–4 week board-ready output, before you commit.

  • Risk-scored map of every voice-triggered workflow inside the bank.
  • Simulated vishing exercises against your contact-centre and treasury teams, with transcripts.
  • Quick-wins list: changes you can make this quarter without a procurement cycle.
  • Structural roadmap: where Real-Time Protection or Detect sit in your existing stack.

Ready to brief your fraud team?

Book a fraud briefingRequest API access