Voice-Clone Fraud Defense Checklist
A realistic AI clone of your CEO’s voice can now call finance and ask for an urgent transfer. Score how well your organisation is defended against voice-cloning / deepfake-voice fraud — the “CEO on the call” and vishing scenario. Work through a weighted checklist of the controls that actually stop a loss: out-of-band verification for payments and credential changes, calling back on a known number, a shared passphrase and a duress code, a “no approvals from voice alone” policy, finance dual control, awareness training and incident reporting. Mark each as in place, partial or not in place and get a 0–100 coverage score, a per-area breakdown, a prioritized gap list and an exportable report. It is a defensive self-assessment: it captures no audio, clones nothing and makes no requests of its own.
A defensive self-assessment of the controls that protect an organisation against voice-cloning / deepfake-voice fraud (the “CEO on the call” / vishing scenario). It makes no network requests, captures no audio and has no voice-synthesis capability. General guidance only — not security or legal advice.
Out-of-band verification
A voice is no longer proof of identity. Confirm high-risk requests on a separate channel.
- Payments and bank-detail changes require out-of-band verificationcritical
Any request to move money or change payee/bank details that arrives by voice (or voicemail) must be confirmed on an independent channel before it is actioned — never approved on the strength of the call alone.
- Call back on a known, pre-stored number — never a number given on the callcritical
A spoofed or cloned caller will supply a number they control. Verification only works if you dial a number from your own directory, not one read out during the suspicious call.
- Credential, MFA and access changes require out-of-band verificationhigh
Voice requests to reset a password, re-enrol MFA, or grant access are a classic deepfake-voice vector against the help desk. These need the same independent confirmation as payments.
Shared secrets & duress
Give people a positive way to prove identity — and a safe way to signal coercion.
- A shared verbal passphrase / challenge exists for sensitive requestshigh
A pre-agreed code word or challenge-response that a cloned voice would not know gives a fast positive check for executive or finance requests, without needing a full callback every time.
- A duress code / safe-signal lets staff flag a coerced requestmedium
Social-engineering calls apply pressure. A duress signal lets an employee comply on the surface while silently alerting security that the request is being coerced or is suspicious.
Policy & approval controls
Remove the single point of failure: one person, one call, one approval.
- Written policy: no money movement or access change is approved from voice alonecritical
The core control. If policy says a voice request is never sufficient authority on its own, a convincing clone of the CEO cannot by itself cause a loss.
- Finance operates dual control / segregation of duties on disbursementshigh
Requiring two independent people to initiate and approve a payment means no single tricked employee — however senior the voice — can complete a fraudulent transfer.
- Elevated / out-of-pattern transfers trigger extra scrutinymedium
Urgency plus an unusually large or novel transfer is the signature of CEO-fraud. Threshold and anomaly checks slow these down for verification.
Awareness & assumptions
People are the sensor. Make sure they know the voice and the caller-ID can both be fake.
- Staff are trained that a familiar voice can be clonedhigh
Deepfake-voice fraud works because people trust a recognised voice. Targeted training for finance, executive assistants and the help desk resets that assumption.
- Caller-ID and inbound numbers are treated as spoofable (never as identity)medium
Caller-ID is trivially spoofed. Staff must not treat a familiar name or number on the display as proof of who is calling.
- Urgency + secrecy is recognised as a red flag, not a reason to skip controlsmedium
"Do it now, tell no one" is the pressure script. Staff should be explicitly empowered to pause and verify exactly when a caller insists there is no time.
Detection & response
Make it easy to report, and know what to do when a suspected deepfake call lands.
- There is an easy, blame-free way to report a suspicious callmedium
Fast reporting lets security warn others, freeze an in-flight transfer, and gather evidence. Friction or fear of blame means incidents are discovered too late.
- An incident playbook covers suspected voice-fraud (recall funds, notify bank, preserve logs)low
When a transfer may already have gone out, speed matters: recall/stop-payment requests to the bank, notification, and evidence preservation have short windows.
Scored entirely in your browser — nothing is uploaded. This tool holds no credentials, captures no audio, makes no outbound requests and has no voice-cloning or synthesis capability of its own; it only scores the answers you tick.
What this tool is
- A weighted 0–100 self-assessment of your organisation’s defenses against voice-cloning / deepfake-voice fraud — the "CEO on the call" and vishing scenario — scored in your browser.
- Covers the controls that actually stop a loss: out-of-band verification for payments and credential changes, callback to a known pre-stored number, a shared verbal passphrase and a duress code, a "no approvals from voice alone" policy, finance dual control, awareness training and caller-ID-is-spoofable assumptions, and easy incident reporting.
- Mark each control as in place, partial or not in place and get a coverage score, a per-area breakdown and a prioritized gap list (critical gaps first), with copy and download.
- Entirely defensive and educational. It makes no network calls, captures no audio and has no voice-synthesis or cloning capability. It cannot clone a voice or place a call.
- Everything runs client-side — nothing you tick is uploaded or stored on a server.
Why this matters
Speech-to-speech voice AI has made a convincing, interactive clone of a known voice cheap and fast to produce. That collapses an assumption a lot of finance and help-desk processes quietly rely on: that recognising someone’s voice is a reasonable proxy for verifying who they are. It is not any more. A caller who can imitate your CFO, invoke urgency, and ask for secrecy is now a realistic attack, and the losses in reported cases have been large.
The good news is that the defenses are not technical mysteries — they are process and policy controls that most organisations can put in place without buying anything. The single most important one is simply deciding that a voice is never sufficient authority on its own, and backing that with out-of-band verification, a known-number callback, and dual control on money movement. This checklist turns those controls into a score so you can see where you stand and which gap to close first. It is defensive guidance only, not security or legal advice.
Frequently asked questions
What is voice-cloning / deepfake-voice fraud?
It is a social-engineering attack in which a criminal uses an AI-generated or AI-cloned voice — often of a senior executive, a known supplier, or a colleague — to pressure an employee into moving money or granting access. The classic form is "CEO fraud" or business email compromise extended to the phone: a convincing voice calls finance, invokes urgency and secrecy, and asks for an urgent wire transfer or a change of bank details. Modern speech-to-speech voice models make the voice realistic and interactive, which is exactly why a voice can no longer be treated as proof of identity. This checklist scores the defensive controls that stop such a call from causing a loss.
How do you defend against a cloned-voice phone call?
The single most important control is policy: no money movement, bank-detail change, or access grant is ever approved on the strength of a voice call alone. Behind that, verify high-risk requests out-of-band — call back on a number already stored in your own directory (never one given during the suspicious call), use a pre-agreed passphrase or challenge-response, and run finance on dual control so no one tricked person can complete a transfer. Train the high-risk roles (finance, executive assistants, the help desk) that a familiar voice can be cloned and that caller-ID is spoofable, and give people a blame-free way to report and a duress signal to flag coercion. This tool walks through each of those controls and scores your coverage.
Why is calling back on a known number so important?
A cloned or spoofed caller controls the number they give you, so "call me back on this number to confirm" is part of the attack, not a defense. Verification only works if you dial a number you already hold — from your directory, HR system or the supplier’s contracted contact record — not one read out or displayed during the inbound request. Caller-ID itself is trivially spoofed, so a familiar name or number on the screen is never identity. Dialling an independently-held number reaches the real person and breaks the attacker’s control of the channel.
What is a duress code and why include one?
A duress code (or safe word) is a pre-agreed signal that lets an employee comply on the surface while silently alerting security that a request is being coerced or looks fraudulent. Voice-fraud scripts apply pressure and urgency, sometimes with a person genuinely under duress. A duress procedure plus a silent escalation path means staff are not forced to choose between confronting a caller and raising the alarm — they can do both safely. It complements, rather than replaces, the out-of-band verification and no-voice-alone controls.
Does this tool clone voices, record audio, or test anything?
No. It is purely a defensive, educational self-assessment. It runs entirely in your browser, makes no network calls, captures no audio and has no voice-synthesis or cloning capability of any kind — it cannot create a deepfake voice or place a call. It only scores the controls you mark as in place and assembles a prioritized gap report from your answers. Treat the output as general guidance, not security or legal advice, and validate your actual controls with a tabletop exercise.
This tool is free and always will be. Would a Pro workspace — saved designs, team sharing and exports — be worth paying for?
Show me Pro →No account needed to use this tool — this just tells us whether Pro is worth building out.