Controls to Prevent AI-Supported CFO Frauds

A finance employee at the Hong Kong office of a London-based engineering firm received a message from someone posing as the group CFO. The message mentioned a confidential transaction, and the employee suspected phishing. Then he joined a video call, saw the CFO and several colleagues on screen, and heard voices he knew. The doubt faded, and he sent 15 transfers worth HK$200 million, about $25.6 million, to five accounts. Hong Kong police said everyone else on that call was a digital fake, as CNN reported.

Looking harder would not have saved him. People flag well-made fake video poorly even when they are told to hunt for fakes, so a defense built on spotting the fake fails at the moment the fake is good. The controls that work keep the payment safe when the voice, the face and the paperwork are all perfect. At the top of the ranking sit a callback to a number you already hold, approval only inside the system of record, two approvers reached on separate channels, and a cooling period for new payees. This article ranks twelve controls from most to least effective and explains how AI builds the fake documents and fake video meetings those controls have to withstand. It is written for controllers, treasury leads, accounts payable managers and the CFOs whose faces get borrowed.

The Hong Kong case was not a one-off. In November 2026, US prosecutors announced an indictment of three people accused of stealing about $129 million from companies between July 2024 and February 2026, using CEO emails, AI-generated executive voices and impostor attorneys, as AP reported through PBS NewsHour. The charges are allegations, and the defendants are presumed innocent. The FBI's Internet Crime Complaint Center counted $3.05 billion in business email compromise losses for 2025, and for the first time it broke out complaints that referenced AI: 22,364 of them, with about $893 million in reported losses across all fraud types, according to coverage of the report.

Turning Fragmented Operational Risks into Decisive Business Action

 

The transition from domain specific risk management to true enterprise risk quantification marks a fundamental shift in how organizations evaluate uncertainty and protect corporate value. Historically, departments have operated within isolated silos. Cybersecurity teams calculate system vulnerabilities, operational units log equipment failures, health and safety groups track physical workplace incidents, and corporate finance tracks liquidity and credit exposures. When presented to executive leadership or the board, these disparate operational domains are often aggregated into chaotic, color-coded heat maps that obscure real financial exposure.

Operational risk quantification solves this fragmentation by establishing a common currency across every corner of the business. By grounding risk analysis in a unified financial loss spectrum and consistent scenario modeling, risk leaders move away from abstract rating systems and deliver concrete decision support that protects corporate balance sheets.

 

How to Plan, Govern, and Build Enterprise AI Agents and AI Architecture

 

AI governance professionals need more than policies. They need to understand how AI systems are designed, where controls belong, and what happens when agents act in the real world.

An AI pilot can look impressive in a demonstration and still fail when a company tries to put it into production. Costs grow, permissions become difficult to manage, outputs vary, and nobody knows exactly who should intervene when something goes wrong.

For risk managers, internal auditors, compliance officers, cybersecurity experts, and IT professionals, these are not abstract technology problems. They are questions of accountability, control design, operational resilience, and business value.

That is why I believe AI architecture deserves a place at the center of enterprise AI governance.

 

Agentic AI in Internal Audit, What to Automate, What to Keep

 

Consider an illustrative scenario, not a real engagement. A mid-sized distributor connects an audit agent to its ERP, its document repository and its ticketing tool. Within a week the agent has pulled invoices, receipts and approvals for 240 purchase orders and produced a tidy memo saying the procure-to-pay control passed. The audit manager reads it, likes it, and asks the only question that matters: which records did the agent open? Nobody can answer.

The short answer to the sequencing problem is this. Agentic AI should first take bounded, repeatable, evidence-based work: a finite set of documents, stable rules, and output you can check against a known result. It must not decide audit scope, judge whether evidence is sufficient, conclude that a control works, rate a finding, or close an issue. Those decisions belong to accountable auditors, and an agent can only prepare the ground for them.

This article is for internal auditors, chief audit executives and the managers and audit committee members who oversee them. After reading it you will be able to sort audit tasks into automate now, automate with review gates, and keep human. You will also have a short list of access rules, evidence requirements and escalation triggers to put in place before any agent touches a workpaper. The question to ask is not whether AI can replace internal auditors. Ask which audit activities are bounded enough to automate safely, and which decisions need independent professional judgment and a named person who answers for them.

 

SOX Controls Over AI

Consider a composite situation, not a real engagement. A controller opens the monthly bank reconciliation, and the matching tool shows 1,250 unmatched items out of 18,400 bank lines. The accounting manager reviewed every item and approved the package. The external auditor asks which version of the matching logic produced that list, whether the bank file was complete when the tool ran, and what the reviewer saw on screen. Nobody can answer. The matching feature arrived in an ERP update that the supplier switched on, and the SOX risk and control matrix still describes a spreadsheet.

If AI influences financial reporting, your organization must be able to show five things: where the AI is used, how it works, that its inputs and outputs are reliable, that changes to it are controlled, and that a human exercised oversight and left evidence of it. Each of those five becomes an auditor question, and this article takes them in the order a SOX program would address them. Treat the five questions as a practical synthesis of current auditor concerns. No official PCAOB or SEC checklist with five questions exists, and I would be wary of anyone who presents one. The article is written for controllers, SOX control owners, and internal auditors who need working answers before fieldwork starts.

You do not need a separate AI compliance program to get there. COSO released guidance on internal control over generative AI in February 2026 that applies its Internal Control Integrated Framework to generative AI, and it names the risks that matter most to financial reporting: prompt-based manipulation, opaque reasoning, model drift, and frequent configuration changes. SEC staff remarks at the December 2025 AICPA and CIMA conference, as summarized in an EY compendium, pointed the same way, with attention on model design, data, human oversight, and the way AI changes IT general controls. What follows turns that direction into an inventory, a control narrative standard, a validation plan, a change register, and a failure path you can build this quarter.

 

AI Governance Frameworks For Risk Managers

 

Corporate environments are adopting autonomous systems at a pace that outstrips traditional oversight mechanisms. Engineering teams ship automated workflows daily. Business units integrate external models into core operations without formal review. The resulting environment creates massive blind spots for risk professionals. You cannot manage what you cannot see, and you certainly cannot audit what you do not understand. The era of treating artificial intelligence as a mere productivity enhancement is over. It is now a foundational component of enterprise architecture. This shift demands a complete rethinking of control environments.

Risk managers, compliance officers, and cybersecurity experts must transition from passive observers to active architects of machine behavior. The theoretical debates about future capabilities no longer matter. The immediate reality involves managing current deployments that process sensitive data, execute financial transactions, and interact directly with customers. Professionals who master this transition will define the next decade of corporate governance. Those who fail will preside over catastrophic compliance failures and severe reputational damage.

This guide provides a direct, actionable blueprint for securing autonomous systems. We will explore five essential pillars of modern risk management. These pillars move beyond basic policy documents. They focus on the practical implementation of controls, the integration of global standards, and the strategic career positioning of governance leaders. You will learn how to validate machine outputs, secure third-party integrations, assign legal accountability, capture institutional context, and manage the hidden costs of automated code generation.

The Risk Management Blueprint: A Practitioner's Guide to Quantitative GRC

 

Risk management has a credibility problem. Not because the profession lacks talent, but because color-coded heat maps, ordinal scoring matrices, and quarterly dashboard reviews were never built to change decisions. They exist to document that a compliance process took place. Executive teams know this. They react accordingly by treating risk departments as corporate overhead instead of strategic assets.

I wrote this book to help my peers turn that dynamic around.

After 25 years leading risk functions and advising executive boards across complex multinational companies, I needed a manual that actually bridges advanced quantitative methods with the daily decisions that determine business outcomes. That drive is why The Risk Management Blueprint hit #9 among the most sold risk management books in the weeks after publishing.

At 867 pages, it gives practitioners a single unified methodology across every major risk domain, covering AI systems, cyber exposure, financial cash flows, sustainability transitions, and human behavior. The framework rests on probability theory, financial modeling, and decision science so you can swap subjective scores for numbers that stand up in the boardroom.

You can preview the first four chapters and access the book here: https://amzn.to/4ciag1F

This is not a textbook. It does not spend the majority of its pages diagnosing what is broken in the profession before gesturing toward improvement in a final chapter. More than 70 percent of the book's total length is allocated to domain applications and advanced analytical infrastructure, meaning the bulk of every page is spent on how to build, calibrate, and apply quantitative and predictive risk models across the decisions that actually shape organizational outcomes.

SOC 2 Is Not Security: A Critical Guide for GRC Managers

Let us be direct. SOC 2 is not a security certification. It is an attestation report issued under AICPA standards in which a licensed public accounting firm expresses an opinion on whether a service organization controls met the selected trust services criteria. The report does not declare that your product is safe. It does not certify that your penetration test was rigorous. It does not prove that your attack surface is small or that your customers are protected from a breach. It states that management described a system, selected criteria, asserted controls, and the auditor found those controls suitably designed and, for a Type II report, operating effectively during the specified period.

That distinction matters more than many teams are willing to admit. The report can create a polished artifact that procurement teams accept, but the underlying controls may still be thin, poorly scoped, or disconnected from the technical reality of the product. The phrase SOC 2 is the audit version of trust me bro became popular for a reason. When the PDF is stronger than the security program, the market starts rewarding documentation rather than operational resilience. As a GRC leader, your job is to reverse that sequence. Build the controls, operate them, collect evidence continuously, and let the SOC 2 report fall out as a byproduct rather than serving as the starting point.

This guide walks through the exact gaps that make SOC 2 incomplete as a security signal, how to read a report without wasting your review cycle, how to build a security-first program that makes SOC 2 the output, how to use continuous assurance strategically, and how to govern vendors that hand you a SOC 2 report and expect instant approval. The focus is practical because the risk owner in the room rarely needs another abstract debate. You need a method for using SOC 2 as one piece of evidence among many.

The article is intended for a global audience. SOC 2 is a US-origin AICPA attestation product, but it is now exchanged across borders by cloud providers, software vendors, data processors, and AI product teams. It often sits alongside ISO/IEC 27001, NIST Cybersecurity Framework, CIS Controls, GDPR, HIPAA, NIS2, and emerging AI governance obligations. That means the underlying discipline remains the same. Understand the limitations, own the controls, and never outsource your risk judgment to a report.

 

The Quantitative Revolution In Enterprise Risk Management

Traditional risk management has reached an inflection point where intuition and qualitative heat maps no longer suffice for navigating complex, interconnected business environments. The modern governance, risk, and compliance director faces a paradox: organizations generate more data than ever before, yet decision makers remain plagued by uncertainty about the very risks that could derail strategic objectives. This gap between information availability and decision quality stems from reliance on uncalibrated expert judgment, measurement of irrelevant variables, and risk models that violate fundamental mathematical principles. The solution lies not in abandoning human expertise, but in rigorously calibrating it through quantitative methods that transform subjective opinions into defensible, mathematically sound probability assessments.

Organizations that master these quantitative techniques gain a decisive competitive advantage. They allocate capital more efficiently by focusing measurement budgets on variables that actually influence decisions. They avoid catastrophic failures by identifying cascade risks and common-mode vulnerabilities before they materialize. They build organizational resilience through models that reflect physical reality rather than statistical convenience. This transformation requires risk professionals to develop new competencies in probability theory, information economics, and computational modeling. The following techniques represent the distilled wisdom of decades of research in decision science, behavioral economics, and quantitative risk analysis. Each method addresses a specific failure mode in traditional risk management, providing practical tools that GRC directors can implement immediately to elevate their organization's risk maturity from descriptive to predictive to prescriptive.

Machine Learning Predictive Risk Modeling for GRC Professionals

AI Use Cases for Risk Management

Machine learning fundamentally transforms risk management from a reactive, sample based discipline into a proactive, population wide surveillance system. The traditional operational model, where risk professionals manually review periodic samples, apply static heuristic rules, and generate retrospective reports, cannot scale to match the velocity, volume, and complexity of modern business transactions. Machine learning enabled systems continuously monitor entire populations of transactions, access requests, supplier relationships, and control events. These systems identify subtle patterns and emerging risks that consistently escape rigid rule based systems. This paradigm shift does not eliminate the need for human expertise. Rather, it repositions risk professionals from data processors to strategic decision makers who focus their judgment on exceptional cases, ambiguous signals, and high consequence approvals. Organizations that successfully implement this model achieve what was previously impossible. They gain comprehensive risk visibility without proportional increases in headcount, enabling the risk function to scale with business growth rather than becoming an operational bottleneck.

The integration of machine learning into governance, risk, and compliance frameworks aligns directly with the core principles of ISO 31000, which emphasizes that risk management must be dynamic, iterative, and responsive to change. Static controls are inherently blind to novel threats and evolving business environments. By embedding predictive analytics into the risk management lifecycle, organizations transition from merely documenting historical failures to actively preventing future exposures. This requires a fundamental rethinking of the risk operating model. The strongest operating model does not seek to replace the risk professional. Instead, it automates the predictable, prioritizes the unusual, and reserves human judgment for material, ambiguous, or consequential decisions. This symbiotic relationship between human expertise and machine scale forms the foundation of modern, resilient risk management.

  

Hiring a Chief Risk Officer: The Interview Questions That Reveal Judgment (With Good and Bad Answers)

Chief Risk Officer hires fail quietly. Not on day one. Not even in the first six months. They fail around month fourteen, when the board and the executive team realise the person they hired can build a risk report but cannot challenge a portfolio manager who is technically within limits but building a position that could unravel the firm.

That is a very expensive lesson.

This guide covers the full hiring process, from mandate definition through structured interviewing to onboarding. It includes specific questions, what good answers look like, and what weak answers reveal. The goal is to help you hire a CRO who makes the firm better at taking risk intelligently, not just one who documents it carefully.

 

AI Isn't Coming for GRC Jobs. It's Coming For The Manual Review Part of Every GRC Job

Here's the uncomfortable part nobody says out loud in a GRC conference room. AI is not replacing risk managers, compliance officers, auditors, cyber teams, controllers, or sustainability experts. It's replacing the manual review work that used to justify half of those job descriptions. What's left after that work disappears is judgment, and judgment is either your biggest career asset right now or the skill you never actually built because the manual work always came first.

Every one of these six professions is being pulled through the same transformation at the same time, just wearing different clothes. Risk teams are using AI to process larger volumes of exposure data faster than any analyst could by hand. Audit is automating the routine testing that used to eat most of fieldwork season. Cyber teams are automating alert triage and first-line response. Compliance is watching AI surface policy conflicts across thousands of documents in the time it used to take to review one contract. Controllers are automating reconciliations and close procedures. Sustainability teams are automating ESG data extraction and disclosure drafting.

None of that is a headcount story on its own. It becomes one for the people who don't adapt, because the professionals who can validate AI outputs, challenge exceptions, and decide where a human still has to sign off are becoming the only ones a board actually needs in the room.

 

S/4HANA Role Redesign: Fix Segregation of Duties Before Go-Live or Pay the Audit Bill After

 

Why Privilege Creep Kills SAP Migrations Before the First Production Transaction Runs

Your migration to SAP S/4HANA is six months out. The project team is focused on data migration, Fiori tile configuration, and cutover planning. Meanwhile, 847 user roles built across eight years of organizational changes, job transfers, emergency firefighter access, and M&A integrations are being lifted wholesale into the new system. Nobody has reviewed them. Nobody has mapped them against the new S/4HANA authorization model. And your external auditors are already asking for the SoD conflict report.

This is the standard failure mode. And it is expensive to fix after go-live.

Migrating unremediated ECC roles into S/4HANA production does not just inherit old access risk. It amplifies it. S/4HANA's simplified data model, new Fiori authorization objects, and transaction replacements create net-new SoD conflicts from role content that was previously clean.

This article gives you the technical remediation workflow to stop that from happening. It covers the SAP-native tools, the sequencing logic, the role design architecture that prevents re-accumulation, and the automated tooling that makes the process viable at enterprise scale.

 

SR 26-2 Is Here: The 2026 Model Risk Guidance That Finally Gives Validators Teeth

On April 17, 2026, the Federal Reserve, the FDIC, and the OCC (collectively, "the agencies") issued SR Letter 26-2, which replaces prior model risk management guidance, the SR 11-7 issued in 2011. This update refines supervisory expectations regarding how banking organizations should calibrate their model risk management frameworks. The guidance is most directly applicable to institutions with total assets exceeding $30 billion, though smaller institutions with complex modeling activities are advised to consider its principles.

The guidance formally excludes simple arithmetic calculations, deterministic rule-based processes, and notably, generative artificial intelligence and agentic artificial intelligence models from the definition of a model. However, the agencies explicitly state that traditional statistical, quantitative, and non-generative artificial intelligence models remain within scope. The primary audience is organizations with over $30 billion in assets, reflecting a tailored supervisory approach that recognizes the lower inherent risk profiles of most community banking institutions.

 

How to Stop Producing Risk Registers Nobody Uses

Enterprise Risk Management programs fail in the same quiet way. They produce polished registers, colorful heat maps, and quarterly reports that look impressive in board packs. Then the organization makes its next major capital allocation, acquisition, or vendor choice using a single-page summary with one projected number and zero reference to the risk framework that consumed thousands of hours to build.

I've watched this pattern destroy the credibility of risk functions across industries. The risk team works hard. Stakeholders get interviewed. Likelihood and impact get scored. And none of it touches the actual decisions that determine whether the organization wins or loses. The gap between risk reporting quality and decision quality is where ERM programs go to die.

This article addresses that gap directly. It provides a stage-by-stage implementation approach for building an ERM program that changes how your organization decides, plans, and allocates resources. Every recommendation comes from field-tested practice, not theory. If your ERM program currently produces documents that live in SharePoint between annual reviews, this post shows you how to fix that.

 

Skills for Compliance Officers, Risk Managers, and Auditors

7 Career Capabilities That Will Separate Compliance Officers Who Thrive in 2026 From Those Who Get Replaced by Algorithms

ING just announced 1,250 job cuts in its compliance operations. ABN Amro plans to replace 35% of its AML division with AI. The Dutch audit office published a report questioning whether the €1.4 billion the banking sector spends annually on anti-money laundering checks actually produces effective outcomes.

Read that last sentence again. The government auditor is asking whether the entire manual compliance model works.

This is not a future scenario. This is happening now, across multiple banks, in one of Europe's most regulated markets. And it raises a question that every compliance officer, risk manager, and internal auditor should be asking themselves today: if my primary value comes from executing manual processes that AI can do faster and more consistently, what exactly is my professional future?

The answer depends entirely on skills. Not certifications. Not years of experience. Skills.

I have spent the last fifteen years working with compliance functions across financial services, industrials, and technology companies. The pattern I see repeating is consistent: the professionals who can quantify risk, challenge AI outputs, and translate regulatory complexity into financial terms the business can act on are becoming more valuable every quarter. The ones who built careers around checklist execution, manual alert processing, and qualitative risk scoring are watching their roles disappear. Sometimes gradually. Sometimes overnight.

This post identifies the seven skills that will define professional survival and advancement in compliance, risk, and audit roles through 2026 and beyond. Each one is grounded in what I see organizations actually hiring for, paying premiums for, and struggling to find.

 

How to Use Large Language Models Securely in Risk Management, Compliance, Cybersecurity, and Audit

 

A compliance officer asked an LLM to analyze a vendor contract for GDPR obligations. The prompt included the full contract text. The contract contained employee names, personal email addresses, salary data from an embedded compensation schedule, and a confidential arbitration clause. All of it went into a third-party API. The compliance officer received a helpful analysis. The organization received a data privacy incident.

Nobody planned for this. The compliance officer was doing good work. The tool produced a useful output. And the organization now had regulated personal data sitting in an external system with no data processing agreement, no retention controls, and no way to request deletion.

That is the paradox of LLMs in GRC. The same capability that makes them powerful for regulatory analysis, risk assessment, and audit automation makes them dangerous when deployed without guardrails. An LLM will process whatever you feed it. It does not distinguish between public regulatory text and confidential personal data. It does not know that the regulation it cited does not exist. It does not understand that the risk score it generated was influenced by training data biases that systematically underweight emerging market vendors.

 

AI for GRC: 10 Use Cases Every Risk and Compliance Team Can Deploy in 90 Days

A compliance analyst at a mid-tier financial institution spent 14 hours last week reading regulatory updates. She flagged three items as potentially relevant to her business. She missed two others that directly affected the firm's cloud outsourcing arrangements. One of those triggered an enforcement action against a peer institution six weeks later.

That story repeats across thousands of GRC teams every week. The volume of regulatory change, vendor risk signals, control evidence, and incident data has exceeded human processing capacity. Not because the people lack skill. Because the volume is physically impossible to cover manually with the rigor the work demands.

AI changes this equation. Not by replacing human judgment, but by compressing the time between a risk signal appearing and a qualified human evaluating it. The 10 use cases in this post are not theoretical. GRC leaders at financial institutions, technology companies, and manufacturing firms are running three to five of these today, cutting manual hours by 30-60% while improving coverage across the full risk population.

Each use case includes the practical workflow, the authoritative framework it maps to, and the implementation path you can follow starting this week.

 

How to Build a Control-Conscious SAP S/4HANA Implementation

Every T-Code, Table, and Design Technique You Need Before Go-Live

The most expensive SAP S/4HANA audit findings are the ones discovered after go-live. Every one of them.

I have watched organizations spend $200,000 remediating a segregation of duties problem that would have cost $5,000 to configure correctly during the explore phase. I have seen a chart of accounts redesign triggered by a post-implementation audit finding that required a partial reimplementation. And I have seen implementations where the system integrator delivered exactly what was specified, but the specifications never included internal controls, so the organization went live with a perfectly built system that had no preventive controls over vendor payments.

These situations are avoidable. Every single one.

This post walks through the complete methodology for building controls into your SAP S/4HANA implementation from day one. It covers the control-conscious implementation team structure, the control design framework with specific T-codes and table references, the audit involvement model, and the SDLC controls that protect the implementation itself. If your implementation is already underway, skip to the section covering your current phase. If you are planning your next implementation, read everything.

 

AI Governance Essentials

Why AI Deployment Now Requires Governance, Not Just Engineering

Artificial intelligence deployment has moved well beyond a technical exercise. What was once framed primarily as model development, application hosting, and iterative improvement now sits squarely within the remit of governance, risk management, compliance, and internal audit. That shift reflects regulatory change, market expectations, and a more mature understanding of how AI systems create both value and exposure across the enterprise.

The original draft focused heavily on product iteration, deployment mechanics, and general technology trends. Those topics matter, but in a professional GRC context they are incomplete unless anchored in accountability, risk ownership, control design, performance monitoring, and assurance. AI systems are not governed effectively merely because they are deployed through modern engineering practices such as CI/CD or MLOps. They require a structured governance model that aligns with recognized frameworks such as ISO/IEC 42001, NIST AI RMF 1.0, COSO Enterprise Risk Management, the IIA Global Internal Audit Standards, and applicable legal obligations including the EU AI Act, privacy laws, sector regulations, and financial reporting requirements where AI affects significant processes.

A practical governance perspective begins with one foundational point. AI is not a single risk category. It is a capability that can introduce, amplify, or obscure multiple risk types at once, including operational risk, model risk, legal risk, compliance risk, information security risk, privacy risk, conduct risk, third-party risk, and reputational risk. In many organizations, this is where governance breaks down. The enterprise treats AI as an innovation program when it should also be treated as a governed business capability subject to the same rigor applied to other material systems and processes.

This distinction matters because controls that are adequate for conventional software may be insufficient for AI-enabled systems. A deterministic business rule can usually be traced to fixed logic. A machine learning model may change behavior as a result of retraining, data drift, feature changes, prompt changes, or vendor model updates. A generative AI application may also produce variable outputs for the same input. For GRC professionals, that means control design must account for non-determinism, data dependency, explainability constraints, and lifecycle volatility.

Effective AI governance therefore starts with a simple but often neglected question. What decision, action, or business process is the AI system influencing, and what is the consequence if it fails or behaves unexpectedly? This framing is more useful than beginning with the underlying algorithm. It allows leaders to assess impact on customers, employees, financial reporting, safety, privacy, regulatory obligations, and organizational objectives.