Why the Usual Program Fails Here

Deepfake-enabled social engineering works by attacking the trusted connection between your people and your tools. Adversaries go after the seams: the handoffs where a person makes a decision, a process is meant to catch it, and technology passes the call to a human under pressure. Those seams are where defenses are least hardened, and that is precisely why attackers aim for them. The attacks that cost real money in the last two years share a signature: the controls that should have held at that junction were the weakest link, and they gave way. Arup's finance team wired $25.6 million across 15 transfers after a video call with a synthetic CFO and colleagues.⁸ MGM took a roughly $100 million hit to a single quarter after one vishing call to a help desk.⁹ In both cases the defense failed where people and tools meet, because that connection is the part of the system organizations harden last.

That is also why awareness training works best with a partner. Good training builds essential knowledge and instincts at the user level, and it is a foundation worth investing in. What it isn't designed to do on its own is tell you whether your process holds, whether your technology hands off safely, or whether all three survive contact together under real adversarial pressure. Completion rates and quiz scores confirm that training was delivered and understood; adversarial testing confirms that it changed behavior when it counts. The playbook below adds that second half, and it starts where every credible security program starts: knowing your actual risk.

You cannot defend what you have not measured, and you cannot measure what you have not simulated. Before you buy a tool, write a policy, or brief the board, you need a defensible answer to one question: where would a real adversary get through, and what would it cost when they did? Most programs cannot answer that. They can report on training completion and email click-through, which measure a narrow slice of the phishing surface and nothing of the infiltration surface. Understanding your risk means replacing assumption with evidence across four dimensions.

Map Your Infiltration Surface, Not Just Your Phishing Surface

A phishing program measures whether people click links in email. The threat that matters now is infiltration: it begins with a compromised or fabricated identity and progresses through your trust architecture: your hiring pipeline, your finance approval workflow, your IT help desk, your vendor and payment-change processes. Inventory those pathways explicitly. A North Korean actor was hired into a U.S. cybersecurity firm through its own hiring pipeline, passing four video interviews and background checks on a stolen identity, detected only after the fact when endpoint monitoring flagged malware on the issued laptop.⁷ Your hiring process is an attack surface. So is every workflow that moves money or grants access.

Identify Your High-Exposure Roles and Processes

Risk is not evenly distributed. Finance and treasury, IT help desk and identity administration, HR and talent acquisition, and executive-adjacent roles carry disproportionate exposure because they can move money, grant access, or authenticate identity. These are the roles an adversary targets first, and they are the ones a generic awareness campaign serves worst. Name them, rank them by the damage a successful compromise would cause, and treat that ranking as the backbone of everything downstream.

Separate Compliance Documentation From Validated Resilience

You almost certainly have documented controls for high-risk transactions: secondary approval, callback verification, passphrase procedures. The question is not whether they exist on paper. It is whether they execute under pressure when a convincing synthetic authority figure is on the line demanding speed. Documentation satisfies an auditor. It does not satisfy an attacker. Until those controls have been stress-tested by someone whose job is to defeat them, you have a policy, not a proven defense, and you should represent it that way to your board.

Establish an Independent Adversarial Baseline

The most useful number you can bring to a board is a current, independent measurement of how your people and processes actually perform against the tactics adversaries are using right now, verified by a third party rather than self-reported. An adversarial baseline gives you a defensible starting point, a way to prioritize spend against real exposure, and a metric you can trend over time to prove the program is working. Without it, every downstream decision is a guess.

• If a synthetic executive appeared on a video call tomorrow and directed a seven-figure transfer, would your secondary-approval process actually stop it, and how do you know?
• Which of your roles and processes represent the highest infiltration exposure, ranked by potential impact?
• Has your identity-verification and hiring workflow ever been tested against a fabricated candidate?
• When was the last time your help desk's caller-verification procedure was defeated on purpose by someone you hired to try?
• Do you have a measured baseline of how your people respond to voice, video, and multi-channel pressure, or only email click rates?
• Can your employees not just recognize a suspected deepfake, but act on a defined escalation path before damage is done?

Once you know where your exposure sits, validate it the way an adversary would exercise it: all three layers at once. Testing any single pillar in isolation reproduces the exact blind spot the attacks exploit. A single-vector email simulation answers whether employees click suspicious links. It tells you nothing about whether they'd authorize a wire transfer for a deepfake executive, whether your finance team's secondary approval actually fires under pressure, or whether an IT help-desk impersonation on Teams would yield a remote-access grant. Multi-vector, multi-stage simulation, sequencing email, voice, SMS, chat, and video into one coordinated campaign, is the only way to measure the coupled system that real attacks target.

Process and procedure are the most effective defense against this threat class, and the least tested. This is where you get the highest return on effort. Enforce out-of-band verification on every high-value transaction and every payment-detail or banking-change request, using a channel and contact established independently of the request itself. Require callback to a number on record, never a number supplied in the request. Implement passphrase or challenge-response verification for sensitive help-desk and executive requests. Then have each of these controls defeated deliberately, under realistic pressure, so you learn where documentation and execution diverge before an adversary does. MGM's help desk had a process; it failed under a ten-minute phone call.⁹

The visual tells are already unreliable, and real-time generation quality is closing what remains. Advice to watch for lip-sync errors and unnatural blinking is a losing strategy; UNC1069's deepfake lure in early 2026 was convincing enough that the victim engaged with it as authentic.⁵ Training builds the knowledge, and experience under realistic conditions is what turns that knowledge into reflex. Use your Step 1 findings to focus your training and simulations on your highest-exposure roles, pair detection with the response you want (verify, escalate, report), and confirm the conditioning held by re-testing under pressure alongside tracking completion.

Recognition without a response path still ends in loss. Your people need to know exactly what to do the moment they suspect a deepfake or a manipulated request: a defined escalation chain, a verification workflow they can invoke without fear of overstepping, and a low-friction reporting channel. Make reporting a suspected synthetic-authority request a rewarded action, not a career risk. The organizations that survive these attacks are the ones where a stressed employee has somewhere to turn other than the instruction in front of them.

Human and process resilience decay. New hires arrive, roles change, tradecraft evolves. Treat adversarial testing as a recurring control, not a one-time project: re-baseline on a defined cadence, remediate the specific gaps each engagement surfaces, and trend the results over time. That trend line is what you bring to the board: not just "we delivered training to 98% of staff," but "our measured exposure in finance dropped from X to Y after we hardened and re-tested the approval process." That is the language that funds the program.

How Breacher.ai Supports This Playbook

Breacher.ai is an independent security research firm led by practitioners. We complement awareness training rather than replace it, and we are the only firm that tests people, process, and technology together using the tactics real adversaries use. Our OSES™ Platform (Orchestrated Social Engineering Simulations™) is purpose-built to give you the Step 1 baseline and the ongoing measurement the rest of this playbook depends on: adversarial assessment across the full attack surface, designed and analyzed by people who have run real incident response and built real detection programs. We have no financial interest in the remediation that follows, which is exactly why our finding is worth trusting.

• Infiltration and Hiring-Pipeline Assessment We test the full infiltration chain, fabricated identities moving through your onboarding and access workflows, not a suspicious email. This is how you close the gap that put a state actor inside a security firm.
• Process Control Validation Under Adversarial Pressure We stress-test your secondary approval, callback, and passphrase procedures against a convincing synthetic authority figure to show you where documentation and execution diverge.
• AI-Powered Vishing and Multi-Channel Orchestration Live voice, video, SMS, chat, and email sequenced into one coordinated campaign, engineered with OSINT-derived context, to measure the coupled system the way an adversary exercises it.
• Independent Baseline and Board-Ready Reporting A defensible starting metric and a repeatable measurement you can trend over time and translate directly into risk language for your board and audit committee.