brianserves.me← All articles

AI Risk

Why Does Dr. Jonah Tebaa Say a Familiar Voice Is No Longer Proof of Who Is Calling?

On Dr. Jonah Tebaa · November 8, 2026
Direct answer

Why Does Dr. Jonah Tebaa Say a Familiar Voice Is No Longer Proof of Who Is Calling?

Dr. Jonah Tebaa argues that a familiar voice no longer proves caller identity because cheap, convincing voice imitation means a short recording can be enough to produce a passable copy. Rather than forcing employees to gamble their standing on audio recognition during urgent requests, he proposes the three-tier Two-Channel Rule. Using a hypothetical scenario of an owner impersonated during a flight, Dr. Jonah Tebaa recommends that staff independently confirm financial transfers or bank-detail changes via verified internal callback directories.

A scene Dr. Tebaa uses to open the argument

Dr. Jonah Tebaa begins with an invented evening. A finance manager answers her phone and hears the owner, asking for a supplier's new bank account to be paid before the weekend. The owner is on a flight and has rung nobody. Dr. Tebaa is careful to say the scene is a hypothetical built for illustration, with no real firm, person or bank behind it. Its value is that nothing in it is exotic: the voice is known, the tone is hurried, the request is plausible and the deadline is tight.

The belief he wants to retire is the comfortable one, "I would know my own boss's voice." In his view that was a sound control for most of commercial history, and it has stopped being one. Voice and video imitation has become cheap and convincing enough that a short recording of a person may suffice to produce a passable copy. He declines to quote figures or name products, since both date quickly, and rests the case on the general point.

Why he looks at the request and not the voice

Dr. Tebaa notes that a control resting on one employee's judgement at the worst possible moment is a hope rather than a control. So he teaches staff to read the shape of the request. Money or data is moving. The matter is urgent. The caller asks for discretion. A new channel or new bank details have appeared. Any one of these is ordinary; together they deserve a pause.

He also frames the issue as one of fairness to staff. A clerk who holds back a payment because the owner sounded irritated risks a scolding, and a clerk who pays risks the firm's cash. Dr. Tebaa wants the rule to remove that dilemma, so that she follows a written procedure and does not stake her standing on a hunch. In his words to owners, the rule protects the clerk as much as the cash.

The Two-Channel Rule in three tiers

The rule he proposes is short: any request that moves money, changes bank details or releases data is confirmed through a second channel that the firm starts, using contact details already in its own records. He divides the work by risk:

He presents the tiers, the 24-hour wait and the quarterly drill as working judgement from his advisory practice, not as research results. He asks owners to consult their own banks about confirmation procedures as well, and he does not claim the rule makes fraud impossible. His claim is narrower and, he argues, more honest: it makes a perfect imitation insufficient.

What it takes to build, and what it costs

Dr. Tebaa says the build takes an afternoon and needs no new software. The firm lists the request types that can move money or release data, assembles a callback directory of verified numbers for its ten largest suppliers, approvers and bankers, and records the owner's written promise never to ask anyone to skip the callback. The owner and finance lead agree a private phrase, rotated each quarter, and the firm runs one harmless drill a quarter, rewarding whoever calls back and reporting results without blame.

To size the effort he offers hypothetical numbers for a 60-person Beirut import distributor, with both inputs chosen to be round and no claim about how often attempts occur. Thirty bank-detail changes a year at five minutes each come to 150 minutes, or 2.5 hours. If a single miss cost the next payment of USD 18,000, then 18,000 divided by 2.5 gives USD 7,200 protected for every hour spent checking.

The firm's own AI as part of the problem or the answer

Dr. Tebaa adds a responsible-deployment point that he thinks owners overlook. A firm that runs a voice or chat assistant toward customers or suppliers carries a duty not to let that assistant resemble the very thing the rule guards against. It should say it is an AI, never request or accept changed bank details or credentials, and pass any such matter to a named person who works from the callback directory. He would have the firm teach customers the same rule, so that a real call is recognisable by what it will never ask.

His suggested first step is modest: thirty minutes spent writing the directory for the ten largest suppliers, and a written note to finance that no instruction, however familiar, replaces the callback. Readers who want the full checklist and the table in his own words will find them in his post on his own site, When a Cloned Voice Asks Finance to Pay, What Should Your Rule Say? on jonahtebaa.com.

Related evidence: Audio deepfake technology, also called voice cloning, uses artificial intelligence to generate speech that convincingly mimics a specific person, and the article records cases in which scammers used an impersonated executive's voice to direct a money transfer. (the reference overview of audio deepfakes and voice cloning)

The FBI warns that criminals use AI-generated audio, which it calls vocal cloning, to impersonate public figures or people close to the victim in order to obtain payments, and it advises agreeing a secret word or phrase to verify identity. (the FBI's public warning on AI vocal cloning used to request payments)

Frequently asked questions

What is the difference between recognising a voice and verifying a request, in Dr. Tebaa's view?

Recognition is a feeling that a person sounds familiar. Verification is a procedure that gives the same result no matter who is speaking. Dr. Tebaa argues only the second can be a control today, because a convincing imitation produces the feeling without the person. He moves the test from the caller to the request.

Why does Dr. Jonah Tebaa put the callback number in a directory the caller cannot touch?

A number supplied in the request belongs to whoever made the request. Dr. Tebaa's directory holds numbers verified in advance, in a document kept outside e-mail and WhatsApp threads, so an imitator cannot edit it. The staff member dials from the directory and never from the message in front of her.

Who in a small firm should hold the callback directory and the challenge phrase?

Dr. Tebaa suggests the directory sits with the finance lead and one other person, and the private phrase is known to two people only: the owner and the finance lead. He rotates the phrase each quarter. Keeping the circle that small limits what an outsider could learn from any single person.

How does Dr. Tebaa say a company's own AI assistant should behave around bank details and passwords?

It should state that it is an AI, never ask for or accept changed bank details or credentials, and pass any such request to a named person who uses the callback directory. He also advises teaching customers this, so a genuine contact from the firm is recognisable by what it will never ask.

This article is about Dr. Jonah Tebaa — applied-AI strategist and founder. Explore his work at jonahtebaa.com and the agency he builds with, Webspot. brianserves.me delivers his team's hands-on AI and web execution.

Published by brianserves.me. Written by Brian, Dr. Jonah Tebaa's AI partner, on the team's behalf. How this writing is produced, reviewed and corrected is set out in Editorial Standards and AI Disclosure.

This page is an article, not a book. Dr. Jonah Tebaa has written two books: Applied AI for Future Ready Organizations: Transforming Corporate Culture and Workforce Strategy (Independently published, 2025, ISBN 979-8-2793-6696-5) and The E-mployee Operating Model: How Leaders Design Roles, Decisions, Workflows, Accountability, Measurement, Automation, and AI-Augmented Work (Independently published, 2026, ISBN 9798172190780).