Skip to main content
AI on the Line: The Voice Phishing Threat, December 1
Blog
Security Awareness Training

Does KnowBe4 Offer Deepfake Training? Evaluating KnowBe4 for Deepfake Protection in 2026

NOVEMBER 20, 202625 MIN READ
Adaptive TeamAdaptive Team

Read summarized version with

Does KnowBe4 Offer Deepfake Training? Evaluating KnowBe4 for Deepfake Protection in 2026

Key takeaways

  • KnowBe4 deepfake training uses an AI agent to generate a scripted video of an organization’s own leader, which administrators assign as a module inside a training campaign;
  • The recommended KnowBe4 deepfake script reveals the fake within the video, so the feature teaches recognition and does not test how employees respond to an unannounced call;
  • KnowBe4 vishing simulations call employees with a pre-recorded or text-to-speech message and record a failure when the employee types the requested digits;
  • As of October 2026, KnowBe4 documentation does not describe an unannounced vishing simulation that delivers a cloned voice of the employees’ own executive or a deepfake video as the live attack, a gap Adaptive Security closes as a deepfake protection platform with unannounced deepfake voice and video phishing simulations;
  • Finance, IT help desk, and executive assistant teams need deepfake phishing training first because they approve payments, reset access, and relay executive requests;
  • Executive impersonation training that clones a leader requires documented consent from that leader and a workforce policy that discloses the program;
  • A verification step that works even when the voice sounds right protects an organization more reliably than training employees to spot audio or video flaws.

Want more from Adaptive in your Google results?

Add Adaptive as a preferred source on Google

KnowBe4 deepfake training, launched in December 2025, uses an AI agent to generate a scripted deepfake video of an organization’s own executive for delivery inside a training campaign. As of October 2026, KnowBe4 documents deepfake training videos using a cloned voice from a source video and a pre-recorded script. However, it does not document an unannounced vishing simulation that delivers a cloned voice of the employees’ own executive or a deepfake video as the live attack.

KnowBe4 deepfake training scripts executive videos while live deepfake simulations remain undocumented

In early 2024, a finance employee at Arup’s Hong Kong office sent HK$200 million, about US$25.6 million, after a video call on which the apparent chief financial officer and colleagues were all deepfake re-creations, and Arup confirmed to CNN in May 2024 that fake voices and images were used. The employee set aside early doubts because the faces and voices matched colleagues he recognized.

Organizations evaluating KnowBe4 deepfake training need to know whether it prepares employees for that unannounced moment or only shows them what a deepfake of their own leader looks like. This KnowBe4 deepfake training evaluation guide covers:

  • How KnowBe4 deepfake training builds and delivers a video of an organization’s own leader;
  • Whether KnowBe4 deepfake simulation and KnowBe4 vishing test employees under live conditions;
  • A deepfake protection platform capability matrix comparing KnowBe4 and Adaptive Security on the same criteria;
  • Which teams need executive impersonation training first;
  • The Pause, Verify, Report Protocol and an annotated deepfake vishing simulation script;
  • Consent requirements for cloning an executive’s voice for a KnowBe4 deepfake or any comparable exercise.

Finance teams that have only watched a deepfake video have never practiced refusing a cloned executive voice under deadline pressure. Adaptive Security runs unannounced deepfake voice and video phishing simulations.

Book a demo

What Is KnowBe4 Deepfake Training and How Does It Work in 2026?

KnowBe4 deepfake training is a feature of KnowBe4 AI Defense Agents (AIDA) that generates a short, scripted deepfake video of a leader from the customer’s own organization for use as cybersecurity awareness training content. KnowBe4 launched the feature on December 15, 2025, and offers it with AIDA or the SAT Advanced tier.

How the Deepfake Training Content Agent Builds a KnowBe4 Deepfake Video

According to KnowBe4’s support documentation, updated on October 13, 2026, the Deepfake Training Content Agent builds each KnowBe4 deepfake video in five steps inside the administrator console.

  1. Name and language: The administrator names the video and selects the language spoken in the source recording;
  2. Source upload: The administrator uploads a landscape .mp4 or .mov recording of the leader, 2 minutes or shorter and no larger than 150 MB, recorded with good lighting, a forward-facing subject, no face obstructions, and clear audio, then accepts a legal authorization prompt;
  3. Scenario selection: The administrator picks a prewritten scenario or writes a custom script of up to 1,000 characters, which passes an automated content check before use;
  4. Voice review: The agent clones the leader’s voice from the uploaded recording, and the administrator listens to the result and can regenerate it;
  5. Video generation: The agent combines the cloned voice with the source video, and the administrator previews the result, saves a draft, or publishes it for use in training campaigns.

Custom scripts can include delivery cues in square brackets, such as [urgent], [calm], or [hesitates], to shape tone and pacing, and those cues count toward the 1,000-character limit. KnowBe4 recommends that custom scripts follow the structure of its prewritten scenarios, which open with a plausible request, reveal that the video is a deepfake, and close with guidance on verifying similar requests.

According to a September 2026 Gartner survey, 41% of 297 cybersecurity leaders surveyed reported at least one social engineering incident involving a deepfake on an employee audio call, and 36% on a video call. The Gartner survey of CISOs on deepfake incidents, published on September 22, 2026, collected those responses between March and May 2026.

KnowBe4 Deepfake Training Requirements, Limits, and Tier Access

KnowBe4 attaches five operating conditions to KnowBe4 deepfake training, each stated in KnowBe4’s support documentation or pricing page as of October 2026.

  • Tier access: The Deepfake Training Content Agent is available with AIDA or the SAT Advanced tier;
  • Video limit: Each account can create up to 25 deepfake videos, tracked on a usage counter in the console;
  • Campaign isolation: A deepfake video must run in a training campaign by itself, and adding one blocks every other content type from that campaign;
  • FedRAMP scope: The feature sits outside KnowBe4’s FedRAMP Moderate Authorization to Operate (ATO), which matters for US public sector buyers;
  • Deletion effect: Deleting a video is permanent and immediately closes any active campaign that uses it.

Administrators track each video on a Deepfakes subtab in the console, where a status column moves from Generating Audio and Audio Ready through Video Ready to Published. A video saved as a draft stays out of training campaigns until an administrator publishes it.

KnowBe4 publishes list prices for both tiers on its website, labeled “Pricing as per May 2026,” as the manufacturer’s suggested retail price (MSRP) in US dollars per seat per month on a three-year term. The table shows those published prices, which KnowBe4 states may change at any time and may vary by region.

Seats SAT Foundation SAT Advanced
25 to 50 $2.40 $3.75
51 to 100 $2.13 $3.40
101 to 250 $1.97 $3.19
251 to 500 $1.80 $2.96
501 to 1,000 $1.63 $2.79
1,001+ Quote Quote

Organizations with more than 1,000 seats receive quoted pricing. Because the deepfake agent requires the higher tier, the SAT Advanced column is the relevant one for any buyer evaluating KnowBe4 deepfake training.

What KnowBe4 Deepfake Training Measures

KnowBe4 lists completion rates, engagement, and improvement over time as the measures for its deepfake training module. KnowBe4’s product page also presents the module as a way to demonstrate behavior change, which remains a vendor claim because KnowBe4 publishes no method connecting video completion to employee behavior.

KnowBe4’s product page states that the module teaches employees how to evaluate audio and video cues and gives guidance on what to do when something feels off. Completion and engagement data confirm that employees watched and finished the KnowBe4 deepfake training video that delivers those lessons. Those measures cannot show how an accounts payable specialist responds when an unannounced call arrives in a cloned executive voice, because the module never places that call.

Security teams that need behavioral evidence can pair the video with a separate test of the verification step itself, such as a callback exercise or an unannounced deepfake vishing simulation. The completion report and the test result then answer two different questions for the same employee.

What does a completion report reveal about how an accounts payable specialist answers a CFO’s cloned voice? Adaptive Security measures that response through unannounced deepfake phishing simulations.

Explore the platform

Does KnowBe4 Offer Deepfake Simulation of Voice and Video Calls?

KnowBe4 offers deepfake training videos and, since a July 30, 2026 announcement, Simulated Vishing phone calls. As of October 2026, KnowBe4 documents deepfake training videos using a cloned voice from a source video and a pre-recorded script. However, it does not document an unannounced vishing simulation that delivers a cloned voice of the employees’ own executive or a deepfake video as the live attack. KnowBe4 deepfake simulation of live voice and video calls featuring an organization’s own executive therefore remains undocumented, even though KnowBe4’s roadmap references AI voice cloning for vishing.

Why KnowBe4 Deepfake Videos Work as Training Content Rather Than a Live Test

Deepfake training content is a lesson in which employees watch a deepfake that the lesson itself identifies as fake. A deepfake simulation is an exercise that delivers synthetic voice or video through a working channel, such as a phone call, video meeting, or SMS thread, and records how the employee responds.

KnowBe4’s support documentation places KnowBe4 deepfake training in the first category. The documentation limits each video to a fixed, pre-recorded script for educational use, and published videos are assigned through training campaigns like any standard KnowBe4 module.

The recommended script structure reinforces that classification, because each prewritten scenario reveals inside the video that the speaker is a deepfake before it explains how to verify similar requests. An employee watching the module learns the outcome before any decision is required, so the video cannot record what that employee would have done.

KnowBe4’s own description supports the same reading. Perry Carpenter, KnowBe4’s chief human risk management strategist, described the feature at its December 15, 2025 launch as providing “a safe, tightly controlled environment for learning.” The same launch materials call the videos simulations, which helps explain why search summaries often describe KnowBe4 deepfake simulation as a testing tool.

Both formats carry value in a cybersecurity awareness training program. Training content builds recognition and executive buy-in, a layer the Adaptive Security guide to deepfakes and security awareness training explains in detail, while a deepfake phishing training exercise measures whether verification happens when the employee does not know a test is underway.

How KnowBe4 Vishing Simulations Work in 2026

KnowBe4 vishing campaigns place automated calls to employees’ phones and play a pre-recorded or text-to-speech message that asks for a PIN or similar number, according to KnowBe4’s Vishing Product Guide. An employee who hangs up without entering digits passes the test, and an employee who types the requested digits fails it.

Administrators can build templates with text-to-speech or by importing MP3 audio files, and results roll into a Vish-prone Percentage and the KnowBe4 Risk Score. Those templates sit beside the wider KnowBe4 phishing simulation program inside the same console, and KnowBe4 describes dozens of prebuilt templates modeled on scenarios such as tax form verification and IT help desk scams.

Campaign settings let administrators place every call when a campaign starts or spread calls randomly across up to four weeks or 30 business days, within business hours set for the account. Calls can display a local caller number for the employee’s country or appear as a private number.

KnowBe4 sets four operating limits on its vishing simulation feature:

  • Call cap: Each user can receive up to 12 answered vishing calls per year, and busy, failed, or voicemail calls do not count toward the cap;
  • Tier access: Vishing requires the SAT Advanced tier or the legacy Diamond tier;
  • Local dialing: Local caller numbers are available in nine regions, including Australia, Brazil, Canada, Israel, the Netherlands, Singapore, South Africa, the United Kingdom, and the United States;
  • Languages: Templates support 26 language and regional-voice options, from Arabic to Cantonese.

KnowBe4 announced Simulated Vishing as a named capability on July 30, 2026, and independent coverage of the July 2026 launch describes local caller ID, realistic personas, and multi-step customizable scenarios. Those three descriptors come from KnowBe4’s launch materials, while the support guide’s keypad response remains the documented method for recording a failure.

According to Verizon’s 2026 Data Breach Investigations Report, the median click rate for mobile-centric vectors, such as voice and text messaging, in phishing simulations is 40% higher than for email. The Verizon 2026 Data Breach Investigations Report limits that comparison to phishing simulation results for voice and text combined, so it measures how often employees fall for tests on those channels rather than how often phone-based cyberattacks succeed.

What KnowBe4 Has Not Documented for Deepfake Vishing Simulation

KnowBe4’s public documentation, reviewed in October 2026, leaves four deepfake vishing simulation capabilities undescribed.

  • Cloned executive voice in a call: KnowBe4’s product roadmap marks Simulated Voice Phishing as completed and describes custom deepfake vishing scenarios built with AI voice cloning, but no KnowBe4 source states that a vishing call can speak in the cloned voice of the customer’s own executive;
  • Two-way conversation: KnowBe4’s July 30, 2026 release describes simulated vishing as text-to-speech calls that engage users in realistic conversations, asking for the kind of data a cyberattacker would request. Published support documentation records a vishing failure when the employee enters the requested digits on the keypad and does not describe speech recognition of free-form employee responses;
  • Unannounced video test: No KnowBe4 source describes a deepfake video delivered through a meeting platform as an unannounced exercise;
  • SMS-to-call chain: No KnowBe4 source describes an SMS step that leads an employee into a follow-up voice call.

Absence from public documentation does not prove absence from KnowBe4’s roadmap, so evaluators should request a live demonstration of each item before a purchase decision. A demonstration also shows how each result would appear in KnowBe4’s reporting and whether a failed call triggers remedial training automatically.

According to Mandiant’s M-Trends 2026, voice phishing accounted for 11% of intrusions, making it the second most common initial infection vector. The Mandiant M-Trends 2026 findings rank only exploits above the phone channel as a starting point for intrusions.

When the only phone test plays a recorded message and waits for keypad digits, cloned executive voices remain untested. Adaptive Security runs unannounced vishing simulations in an executive’s voice.

Book a demo

KnowBe4 vs. Adaptive Security: Comparing Deepfake Protection Platforms’ Capabilities

Compared on documented capabilities as of October 2026, KnowBe4 and Adaptive Security both offer custom executive deepfake training content, and only Adaptive Security documents unannounced deepfake voice and video simulations. The deepfake protection platform comparison uses 10 criteria covering training content, phishing simulation channels, verification practice, and reporting.

How the Deepfake Protection Platform Matrix Compares Both Vendors

The matrix compares KnowBe4 and Adaptive Security as a deepfake protection platform from one angle, training content versus unannounced deepfake simulation, using 10 identical criteria. Every cell reflects each vendor’s own public product pages and support documentation, so both columns are vendor-reported, and a separate Adaptive Security vs. KnowBe4 feature comparison covers the broader product lines.

Criterion KnowBe4 (as of October 2026) Adaptive Security (as of October 2026)
Deepfake awareness content Deepfake and AI cyber threat modules in the KnowBe4 Studios library, plus custom deepfake training videos Training library covering deepfakes, voice scams, and AI phishing, plus custom deepfake personas embedded in modules
Executive likeness and voice cloning Training video built from an uploaded source recording, with a legal authorization prompt Personas built from a photo and voice sample for modules, plus phishing simulations featuring an organization’s executives
Delivery of the executive deepfake Assigned training module whose recommended script reveals the fake Training modules and phishing simulations, with Auto-Phish generating simulations per employee
Voice phishing simulation Automated calls with pre-recorded or text-to-speech audio and a keypad response, capped at 12 answered calls per user per year AI-generated vishing simulations with voice cloning, built from public OSINT and deployed alongside SMS and email
Cloned executive voice inside a vishing simulation Documented as a controlled training video (pre-recorded script, cloned voice); not documented as an unannounced vishing test using the customer’s own executive’s cloned voice Documented: voice cloning simulations and executive impersonation built from public OSINT
Video deepfake simulation Not documented as an unannounced test Deepfake video and audio simulations built from public open-source intelligence (OSINT)
SMS phishing simulation No SMS test type in KnowBe4’s list of simulated phishing tests; smishing is covered through training content SMS phishing simulations
Multi-channel attack chains Multi-step scenarios announced for vishing Coordinated email, voice, SMS, and chat campaigns
Verification-protocol training Script guidance on verifying similar requests Just-in-time micro-lessons when an employee slips, plus training assigned automatically by role or risk score
Reporting and risk scoring Completion and engagement for deepfake modules; Vish-prone Percentage and Risk Score for vishing Per-employee and per-department risk scores, plus executive OSINT dossiers that flag deepfake and phone-channel risk

For showing employees what a deepfake of their own leader looks like, both vendors offer custom training content. For measuring how employees respond when a cloned executive voice or video arrives without warning, Adaptive Security documents the capability and KnowBe4 does not as of October 2026.

Where KnowBe4 Deepfake Coverage Is Strong

KnowBe4 deepfake coverage holds three advantages for organizations already running KnowBe4, based on the vendor’s own documentation.

  • Program integration: The deepfake video runs inside an existing KnowBe4 cybersecurity awareness training program, with up to 25 videos per account and prewritten scenarios that need no scriptwriting, and its completion data sits in the same reports as other KnowBe4 campaigns;
  • Administrator control: Every script is written or selected by an administrator and screened before generation, which keeps the content inside the security team’s approval process;
  • Adjacent content: The KnowBe4 Advanced library of more than 1,000 pieces includes vishing modules and a vishing recognition game called Spot the Vish, which support KnowBe4 deepfake awareness campaigns around the video.

For a customer already on the SAT Advanced tier, adding a deepfake module requires no new vendor, contract, or console. That convenience makes the feature a low-effort way to introduce executive deepfakes to a workforce that has never seen one.

Where the Gaps Sit for Deepfake Vishing Simulation and Video Calls

According to Crane Authentication data in APWG’s Phishing Activity Trends Report, 2nd Quarter 2026, vishing rose 20% from the first to the second quarter of 2026. The APWG Phishing Activity Trends Report attributes its smishing and vishing figures to data from Crane Authentication.

Against that growth, KnowBe4’s documented coverage leaves five gaps for deepfake vishing simulation and video calls.

  • Generic phone voices: Documented phone tests use pre-recorded or text-to-speech audio; KnowBe4’s support guide does not describe calls in the cloned voice of the organization’s own leader, although its roadmap lists AI voice cloning for vishing;
  • Documented keypad response: KnowBe4’s support guide records a failure only when digits are entered and does not document how a verbal disclosure or an agreed payment would be scored;
  • Announced video only: The deepfake video reaches employees as a labeled lesson inside a campaign;
  • Campaign isolation: A deepfake video cannot share a campaign with other content types;
  • FedRAMP exclusion: Public sector buyers that require FedRAMP Moderate coverage cannot count the deepfake feature within it.

Finance approvers and help desk agents feel the first three gaps most, because a cloned leader’s voice is the pretext most likely to reach them. Adaptive Security phishing simulations address those voice and video gaps with unannounced deepfake voice, video, SMS, and email exercises built from an organization’s own executives. Each result then shows what the employee did rather than what the employee watched.

A training video shows employees a labeled fake, while a cyberattacker’s call arrives unlabeled and urgent. Adaptive Security tests the second scenario with deepfake vishing simulations across voice and video.

Take a self-guided tour

Why Live Simulations Are a Must for Deepfake Phishing Training in 2026

Realistic voice and video practice trains employees to refuse familiar voices under time pressure

Deepfake phishing training works best when employees practice the verification step against a realistic voice or video under time pressure. Cyberattackers now pair cloned voices with urgent payment and access requests, and a watched video cannot rehearse the moment an employee must refuse a familiar voice.

How AI Voice Cloning Attacks Reach Employees

AI voice cloning attacks follow a five-stage sequence that moves from public recordings to an urgent request.

  1. Harvesting: Cyberattackers collect public audio and video of an executive from keynotes, earnings calls, interviews, and online videos;
  2. Priming: A pretext email or message tells the target that a call is coming or that a confidential matter needs handling;
  3. Impersonation: The target receives a call or voice note in the cloned voice of the executive;
  4. Process bypass: The cloned voice requests a payment, credential reset, or data transfer that skips normal approval;
  5. Isolation: The caller demands secrecy or moves the conversation to a channel the organization does not monitor.

The first stage costs cyberattackers little, because executives with public roles supply the source audio themselves. Every later stage depends on the target treating a familiar voice as proof of identity.

According to the FBI’s 2025 IC3 Annual Report, over $30 million in business email compromise losses involved AI, with voice cloning noted as a way to request wire payments. The FBI 2025 IC3 Annual Report counts only complaints filed with the Internet Crime Complaint Center (IC3), so the figure understates losses that went unreported.

The fourth stage is where an employee’s decision determines the outcome of AI voice cloning attacks, and the Adaptive Security guide on how to defend against AI voice cloning scams covers the callback defense for that decision point. Training that rehearses the fourth stage protects the organization even when the first three stages succeed.

Deepfake Phishing Training Statistics for 2026

According to IBM’s Cost of a Data Breach Report 2026, deepfake impersonation was the most common AI-enabled cyberattack type at 45%, as published in IBM’s UK release on AI-enabled breaches. IBM published that breakdown in its UK release.

These findings set the scope for deepfake phishing training in 2026. A program that covers only email leaves employees unprepared for the voice and video channels where deepfake impersonation lands, and a program that covers voice through recorded messages alone leaves the cloned-voice scenario untested.

The figures also point to where training effort belongs. Deepfake impersonation concentrates on requests that move money or grant access, so the employees who handle those requests need the most frequent practice. Their results also need to be measured separately from the general workforce.

According to the FBI’s 2025 IC3 Annual Report, its AI section recorded 22,364 AI-related complaints with $893,346,472 in adjusted losses.

Why Recognition Alone Fails Under Pressure

Recognition-based training teaches employees to look for audio and visual flaws, and that approach weakens whenever a deepfake contains no flaw the employee can detect. A verification process that never depends on spotting the fake keeps working as synthetic media improves.

Pressure compounds the problem. A cloned executive voice arrives with a deadline, a reason for secrecy, and the authority of a senior title, and each element shortens the time an employee spends deciding. A rule the employee can cite is therefore more dependable than a judgment the employee must make alone.

According to CrowdStrike’s 2026 Global Threat Report, there was an 89% increase in attacks by AI-enabled adversaries in 2025. As the AI-enabled attacks measured in the CrowdStrike 2026 Global Threat Report scale, a control that holds regardless of how convincing the synthetic voice becomes is safer than one that depends on an employee’s ear.

The NSA, FBI, and CISA information sheet Contextualizing Deepfake Threats to Organizations, published in September 2023, advises organizations to prepare to identify, defend against, and respond to deepfake cyber threats, including a plan to respond and minimize impact. The same guidance treats deepfake detection technology as one measure among several, which places the employee’s verification habit inside the defense rather than outside it. For deepfake phishing training, that guidance means an exercise succeeds when the employee follows the response plan, whether or not the employee ever suspected the voice.

Rehearse the callback step under realistic pressure with Adaptive Security, which delivers voice cloning and deepfake simulations tailored to the roles that approve payments and reset account access.

Book a demo

Which Employees Need Executive Impersonation Training First?

Experience the Adaptive platform

Take a free tour

Finance and accounts payable teams, IT help desk staff, executive assistants and executives, and recruiting and HR teams need executive impersonation training first. Each group can move money, reset access, relay a leader’s instructions, or bring a new person into the organization, which are the actions cyberattackers impersonate leaders to obtain.

The four priority groups map to four distinct actions that a cloned executive voice tries to trigger.

  • Finance and accounts payable: Payment approvals and vendor bank-detail changes,
  • IT help desk: Password and multi-factor authentication (MFA) resets,
  • Executive assistants and executives: Instruction relays, calendars, and travel details,
  • Recruiting and HR: Interviews and onboarding of remote candidates.

Finance and Accounts Payable Teams

Finance and accounts payable teams approve payments and process vendor bank-detail changes, which are the transactions a cloned executive voice most directly targets. Where one employee can release a payment alone, an urgent request from a familiar voice to pay a new beneficiary is all a cyberattacker needs.

According to the FBI’s 2025 IC3 Annual Report, victims reported $3,046,598,558 in business email compromise losses across 24,768 complaints. Business email compromise (BEC) relies on the same payment workflows that finance teams control, whether the request arrives by email, phone, or video.

Executive impersonation training for finance should center on new-beneficiary and changed-bank-detail scenarios delivered by phone and video. A second approver for any new payment destination gives the trained employee a process to point to when a senior voice demands speed, and it makes a single convinced employee insufficient to release funds.

IT Help Desk Staff

According to Mandiant’s M-Trends 2026, vishing accounted for 23% of all cloud-related security incidents. Google Cloud’s Office of the CISO cited the figure on June 11, 2026, in its analysis of vishing as a share of cloud incidents, which describes cyberattackers deploying cloned executive voices during high-pressure help desk calls.

The same Google Cloud analysis identifies speed-based help desk metrics, such as average handle time, as a vulnerability because they push agents to prioritize speed over strict identity verification on identity-sensitive calls. The Adaptive Security explainer on how vishing attacks work breaks these calls down step by step.

Help desk training should establish that no password or MFA reset proceeds on voice recognition alone, regardless of the caller’s seniority or stated urgency. The agent then has a written rule to cite when an executive’s voice insists on an exception.

Executive Assistants and Executives

Executive assistants relay instructions on a leader’s behalf and manage calendars and travel details, so a cloned voice that sounds like their executive carries immediate authority. Executives face the inverse exposure, because their public recordings supply the source audio and their names supply the pretext.

Training for both roles should establish an agreed out-of-band check, such as a callback to a known number, for any unusual request in either direction. Executive impersonation training for leaders also shows them how their own voice could be used against their team.

According to Verizon’s 2026 Data Breach Investigations Report, pretexting reached 6% of breaches as an initial access vector. Pretexting is a social engineering technique in which a cyberattacker invents a scenario, such as an urgent executive request, to persuade the target to act. Executive assistants sit at the center of those scenarios.

Recruiting and HR Teams

Recruiting and HR teams conduct video interviews and onboard remote candidates, which makes them the entry point for synthetic candidates who combine deepfakes with stolen identities to get hired. A hire made on false credentials gains internal system access that no phishing email could deliver.

According to the FBI’s 2025 IC3 Annual Report, almost $13 million in losses involved AI-involved employment scams, and the report notes possible voice deepfakes in online job interviews.

CyberScoop reported in July 2024 that KnowBe4 disclosed hiring a North Korean IT worker who used a stolen US identity and an AI-enhanced stock photo, and KnowBe4’s security team contained the workstation after detecting malware on it. Recruiting teams therefore benefit from identity verification steps that do not rely on a video interview alone.

Adaptive Security builds each deepfake phishing simulation from public OSINT and an employee’s role, so finance approvers, help desk agents, and executive assistants practice their most likely request.

Explore the platform

The Pause, Verify, Report Protocol for Deepfake Vishing Simulation

The Pause, Verify, Report Protocol is a three-step response for any unexpected request involving money, credentials, or access that arrives by voice or video. The protocol works even when the employee believes the caller is genuine, which makes it the behavior a deepfake vishing simulation should rehearse and score.

The Three Steps of the Pause, Verify, Report Protocol

The Pause, Verify, Report Protocol gives employees a three-step response to unexpected voice and video requests

The protocol assigns one action to each step, and the order matters because each step removes a different lever the caller relies on.

  1. Pause: Stop the requested action, regardless of the stated deadline, the caller’s seniority, or how familiar the voice sounds;
  2. Verify: End the call and contact the requester through a phone number or channel listed in the company directory, and require a second approver for any payment or access change;
  3. Report: Send the caller’s number, the time, and the request details to the security team through the approved reporting channel.

The verify step works because a callback to a directory number takes the verification channel out of the cyberattacker’s control. A cyberattacker can spoof caller ID and clone a voice, yet has no way to intercept a callback placed to the executive’s listed number.

The report step extends one employee’s suspicion into protection for the whole organization, because the security team can warn other likely targets before the same caller reaches them. The report also gives the security team the caller’s number and pretext to flag across the organization.

The protocol applies equally to a phone call, a voice note, a video meeting, or a message asking for a callback, because each one requests action before verification. Publishing the protocol beside payment and access policies gives employees a written rule to cite when declining a senior caller.

According to Verizon’s 2026 Data Breach Investigations Report, a human element was present in 62% of breaches, up from 60%. A protocol that depends on a fixed sequence of actions, rather than on judgment about whether a voice sounds authentic, addresses that human element directly.

An Annotated Deepfake Vishing Script With Each Red Flag Explained

Hypothetical scenario. The following script follows a cloned-voice call to an accounts payable specialist through six beats, each annotated with the red flag it contains.

  1. Unscheduled contact: The specialist’s mobile phone rings with a number that displays as the chief financial officer’s, and a familiar voice says, “I need something handled quietly before the bank closes.” The red flag is that caller ID can be spoofed, so a familiar number confirms nothing about who is calling;
  2. Manufactured urgency: The voice says a confidential acquisition payment must clear before the bank’s cutoff in 40 minutes. The red flag is a deadline that leaves no time for the normal approval cycle, which is the purpose of the deadline;
  3. Process bypass: The caller asks the specialist to pay a new beneficiary and skip the second approver “just this once.” The red flag is any request to suspend a control, because the approval process exists for exactly this kind of payment;
  4. Secrecy: The caller instructs the specialist not to copy anyone because the deal is under embargo. The red flag is isolation, because secrecy prevents the colleague check that would expose the fraud;
  5. Channel lock: When the specialist offers to call back, the caller says they are boarding a flight and can only stay on this line. The red flag is refusal of a callback, which suggests the caller cannot be reached at the executive’s listed number;
  6. Authority and flattery: The caller says the specialist is the only person trusted with the transfer. The red flag is flattery paired with authority, a combination that discourages the specialist from questioning a senior leader.

Financial regulators publish comparable indicators for deepfake-enabled fraud. The FinCEN alert on deepfake fraud schemes, issued as FIN-2024-Alert004 in November 2024, lists red flag indicators for financial institutions, such as a customer using a third-party webcam plugin during a live verification check.

Applied to this script, the specialist pauses before entering any payment details, verifies by calling the chief financial officer’s directory number, and reports the incoming number and request to the security team. The cloned voice sounding exactly right counts as a red flag in neither direction, because the protocol never asks the specialist to judge the voice. The specialist follows the same three steps whether the caller is a cyberattacker or the chief financial officer, and the cost of a callback to a genuine executive is a few minutes.

How to Rehearse the Protocol in a Deepfake Vishing Simulation

Security teams can run the annotated script as an unannounced deepfake vishing simulation for the four priority groups, using a cloned executive voice only with that executive’s written consent. The Adaptive Security complete guide to vishing covers the wider set of vishing techniques that scripts like this one draw from.

Scoring should account for whether the employee paused, called back on a directory number, and reported the call, rather than whether the employee identified the voice as synthetic. That scoring keeps the exercise aligned with the protocol the employee is expected to follow. Running the same script against finance, help desk, and executive support staff also shows whether one group follows the protocol less consistently than the others.

A debrief immediately after the call, delivered without penalty, shows the employee which beat should have triggered the pause. Coaching without penalty also signals that reporting a suspicious call, even a late report, is the outcome the organization wants.

Verification protocols that employees have only read about tend to collapse once a familiar voice insists on speed. Adaptive Security tests them inside unannounced calls using an executive’s cloned voice.

Take a self-guided tour

Consent is required in practice, because KnowBe4 makes administrators confirm legal authorization from the person depicted before generating a KnowBe4 deepfake. Any program that clones an executive should also hold that executive’s documented consent, disclose the program in policy, and review applicable law with counsel.

How KnowBe4 Handles Authorization for Deepfake Training Videos

KnowBe4 lists legal authorization from the individual recorded as a prerequisite for every deepfake video, and a pop-up with legal requirements must be accepted before the source upload proceeds. Scripts then pass an automated filter that blocks sexual, violent, hateful, and harassing language.

KnowBe4’s December 15, 2025 launch announcement states that administrators create and approve all of its deepfake simulations, a design choice KnowBe4 presents as ensuring ethical use.

Because the recommended script reveals the deepfake inside the video, viewers learn that the content is synthetic before the module ends. That design covers disclosure to viewers while leaving consent from the executive to the administrator’s authorization step. KnowBe4’s support guide does not describe a consent form, a consent duration, or how long uploaded source recordings are retained, so those terms fall to the customer’s own policy.

Legal Considerations for Executive Voice Cloning in 2026

Two rules shape executive voice cloning in 2026, and how each applies to internal security exercises remains legally unsettled as of October 2026. The analysis that follows is general information rather than legal advice, and organizations should confirm their position with counsel before cloning any executive’s voice.

Article 50 of the EU AI Act, part of Regulation (EU) 2024/1689, requires deployers of an AI system that generates or manipulates image, audio, or video content constituting a deepfake to disclose that the content has been artificially generated or manipulated. The Article 50 transparency obligations have applied since August 2, 2026, and how disclosure timing works for an unannounced internal deepfake phishing simulation remains an open question. Article 50 limits disclosure for evidently artistic, creative, satirical, or fictional works to a form that does not hamper the work, and whether a training exercise qualifies is also unsettled.

In the United States, the FCC ruling on AI-generated voices, adopted on February 2, 2024, and released on February 8, 2024, recognizes AI-generated voices as “artificial” under the Telephone Consumer Protection Act (TCPA), which restricts calls that use an artificial or prerecorded voice. The ruling took effect immediately and gives state attorneys general new tools to pursue voice cloning scams. Programs that place automated vishing simulation calls to employees’ personal mobile numbers should confirm with counsel how the TCPA applies.

According to the World Economic Forum’s Global Cybersecurity Outlook 2026, the share of organizations assessing the security of AI tools rose from 37% in 2025 to 64% in 2026. A voice-cloning tool used for executive impersonation training belongs in that assessment scope like any other AI tool, including the question of where its source recordings are stored.

An Ethics Checklist for Executive Impersonation Training

Six controls keep executive impersonation training defensible to executives, employees, and regulators, and each one can be documented before the first cloned call is placed.

  • Written consent: Each cloned executive signs a consent that defines the scope, the channels, and an end date, and the consent can be withdrawn at any time;
  • Policy disclosure: The acceptable use or security policy states that phishing simulations may use synthetic voices or likenesses of leaders, without revealing timing;
  • Script limits: Scripts avoid personal, medical, family, or emergency scenarios that would cause distress if believed, and they stay within business requests such as payments, access, and data;
  • No punitive outcomes: Employees who fail a cloned-voice exercise receive coaching and face no disciplinary consequence, so the exercise does not discourage reporting;
  • Data handling: Source recordings and cloned voice models are stored securely and deleted when consent ends;
  • Executive briefing: Each cloned executive knows which scenarios are running, so the executive can confirm or deny a request quickly when an employee calls back.

An executive who has signed the consent and knows the scenario can answer an employee’s verification call promptly, which keeps a training exercise from escalating into an incident response. The briefing also lets the executive answer colleagues who ask afterward whether the call was genuine.

An executive requested to approve a cloned-voice exercise should first see why that voice is a target. Adaptive Security’s executive risk intelligence maps each leader’s deepfake and phone-channel exposure using Risk Monitoring.

Book a demo

How to Choose a Deepfake Protection Platform in 2026

Selecting a deepfake protection platform starts with the behavior the organization needs to measure, then weighs five criteria in order of importance for high-risk roles. The final choice depends on whether the goal is awareness of what deepfakes look like or evidence of how employees respond to one.

Five Criteria for Evaluating Deepfake Phishing Training Platforms

According to the UK Department for Science, Innovation and Technology’s Cyber Security Breaches Survey 2025/2026, 22% of businesses tested staff, for example with mock phishing exercises. Testing remains a minority practice among UK businesses before voice and video channels even enter the picture. A deepfake phishing training vendor that tests across channels therefore addresses a gap most of those businesses have not closed.

Five criteria separate deepfake phishing training vendors on the capabilities that matter most for high-risk roles.

  1. Deepfake simulation channels: The vendor delivers unannounced deepfake exercises across voice, video, SMS, and email, matching the channels cyberattackers combine in one campaign. A vendor limited to email or recorded phone messages leaves the cloned-voice request untested;
  2. Executive modeling: The vendor can clone the organization’s own leaders, with consent controls and administrator approval of every script. A familiar executive voice tests the exact trust a cyberattacker exploits, which a generic voice cannot;
  3. Response capture: The vendor records what the employee did after answering, beyond a click or a keypad entry. Callbacks, reports, and refusals are the behaviors that predict whether a payment would have left the account;
  4. Verification rehearsal: Scenarios reward the callback step and reporting, rather than only measuring whether the employee was fooled. Rewarding the correct action teaches the protocol faster than penalizing the mistake;
  5. Risk reporting: Results feed behavioral metrics by role and department instead of completion rates alone. Leadership can then see which teams would release funds to a cloned voice and which would call back first.

Organizations with high payment volumes should weight deepfake simulation channels and response capture most heavily, because those two criteria measure the finance team’s behavior directly. A feature-level view of Adaptive Security compared with KnowBe4 is available on Adaptive Security’s comparison page for teams assessing the broader product lines.

When KnowBe4 Deepfake Training Fits and When It Does Not

The choice between KnowBe4 deepfake training and a vendor that runs unannounced deepfake simulations follows the organization’s goal, its current vendor, and its compliance obligations, as the following table shows.

KnowBe4 deepfake training fits when A live deepfake simulation platform fits when
The organization already holds AIDA or the SAT Advanced tier The goal is measuring how employees behave under pressure
The goal is awareness and leadership buy-in Finance and help desk roles face cloned-voice requests
FedRAMP coverage is not required for the deepfake feature The program needs SMS or video tests

According to IBM’s Cost of a Data Breach Report 2026, one in four malicious breaches was AI-enabled, a 56% increase over last year, with an average cost of $6 million. The IBM Cost of a Data Breach Report 2026 places AI-enabled breaches above its overall global average cost, which raises the stakes of a format that only shows employees a deepfake.

Awareness content and unannounced testing can run side by side. Organizations considering switching from KnowBe4 can begin by piloting unannounced deepfake exercises with finance and help desk staff while existing KnowBe4 modules continue. The pilot results then show whether the existing program’s completion rates matched how those teams behave on a cloned-voice call, which gives procurement a behavioral baseline before any contract renewal.

Completion reports record who watched a deepfake video while revealing nothing about who would pay heed to a cloned executive’s urgent request. Adaptive Security tests that decision unannounced and rolls every interaction into risk scores.

Take a self-guided tour

How Adaptive Security Runs Deepfake Vishing Simulations of Executives

Adaptive Security builds deepfake vishing simulations from OSINT to show how key roles respond

Adaptive Security shows security leaders how finance, help desk, and executive support staff respond when a cloned executive voice or deepfake video arrives unannounced. Adaptive Security’s Phishing Simulations build each deepfake vishing simulation from OSINT and an organization’s own executives, then deliver it by phone, video, SMS, or email.

When an employee falls for an exercise, a just-in-time micro-lesson explains the missed cue on the spot. In Security Awareness Training, an executive likeness can become a custom deepfake persona for awareness modules, so recognition training and unannounced tests run inside one cybersecurity awareness training program.

Interactions roll up into Risk Monitoring and Mitigation, which updates each employee’s risk score daily and shows which teams verify before acting. Executive risk intelligence maps each leader’s deepfake and phone-channel exposure, so the next executive impersonation training cycle starts with the leaders whose voices are most exposed.

Deepfake awareness training content prepares employees to recognize the cyber threat, and unannounced exercises show whether they verify under pressure. Adaptive Security delivers both inside one cybersecurity awareness training program.

Take a self-guided tour

Frequently Asked Questions About KnowBe4 Deepfake Training

Does KnowBe4 Offer Deepfake Training?

Yes, KnowBe4 offers deepfake training through its Deepfake Training Content Agent, which generates a scripted deepfake video of a leader from the customer’s own organization. The agent is available with AIDA or the SAT Advanced tier, and administrators assign the finished video as a module inside a training campaign, where it runs as KnowBe4 deepfake awareness content for employees.

Does KnowBe4 Simulate Deepfake Voice or Video Calls?

As of October 2026, KnowBe4 does not document an unannounced vishing simulation that delivers a cloned voice of the employees’ own executive or a deepfake video as the live attack. Its documentation describes deepfake training videos that employees watch inside a training campaign and automated KnowBe4 vishing calls that use pre-recorded or text-to-speech audio with a keypad response. The deepfake training videos do use a cloned voice from a source video, but with a pre-recorded script delivered inside a training campaign.

Is KnowBe4 Training AI-Generated?

KnowBe4 training is partly AI-generated. The core KnowBe4 Studios library consists of produced content such as videos, games, and live-action series. KnowBe4’s AI Defense Agents (AIDA) generate phishing templates, custom training built from an organization’s policies, and deepfake videos of the customer’s own leaders. Administrators decide which mix of produced and generated content each campaign uses.

Does KnowBe4 Offer AI Training?

Yes, KnowBe4 offers training on AI-era cyber threats and uses AI to deliver it. The KnowBe4 library includes content on deepfakes and shadow AI. KnowBe4 also markets a suite of AI Defense Agents that personalize training assignments, generate phishing simulations, build quizzes from company policies, and create KnowBe4 deepfake training featuring the organization’s leaders.

Is KnowBe4 Deepfake Training Free?

No, KnowBe4 deepfake training requires a paid AIDA subscription or the SAT Advanced tier. KnowBe4 does publish separate free tools, such as a baseline phishing test and an email reporting add-in. Its terms of service also provide time-limited evaluation accounts that let prospective customers review the console before committing to a paid tier.

Can Deepfakes Be Detected by Software Alone?

No, detection software alone is an incomplete defense against deepfake fraud. Detection tools can flag some synthetic audio and video, but US government guidance on deepfake cyber threats treats detection as one measure alongside preparation and response planning. A callback to a known number protects a payment even when both the software and the employee fail to recognize the fake.

How Often Should Deepfake Simulations Run?

No published standard sets a frequency for deepfake simulations as of October 2026. A practical approach ties cadence to role risk, running each deepfake vishing simulation more often for finance approvers and help desk agents than for the general workforce. Exercises should also repeat after a payment process changes or after a cloned-voice incident reaches the organization.

How Many Deepfake Videos Can a KnowBe4 Account Create?

A KnowBe4 account can create up to 25 deepfake videos, according to KnowBe4’s support documentation as of October 2026. Each published video must run in a training campaign of its own, and deleting a video closes any active campaign that uses it. The feature also sits outside the scope of KnowBe4’s FedRAMP Moderate authorization.

How many employees would call back before acting on a cloned executive’s voice? Adaptive Security answers that question with deepfake vishing simulations built from an organization’s own leaders and workflows.

Book a demo

Adaptive Team

Adaptive Team

As experts in cybersecurity insights and AI threat analysis, the Adaptive Security Team is sharing its expertise with organizations.

Get started with Adaptive Security

Human and Agent Security for the AI Era.