AI Model RIsk, Validation and Assurance

AI Model Risk, Validation and Assurance

AI Model Risk, Validation and Assurance equips risk, compliance, model validation, data governance and audit professionals with the skills to independently assess, validate and monitor AI systems. Grounded in Malaysia’s evolving AI regulatory landscape, the programme focuses on ensuring AI systems are reliable, transparent, accountable and compliant. It builds on AI Governance for Business Leaders by moving beyond establishing governance controls to verifying their effectiveness in practice.

DOWNLOAD BROCHURE

- Target Roles: Mid- to senior-level professionals in second and third-line defence functions responsible for validating, auditing, monitoring, or assuring AI systems. This includes Model Risk Management, Model Validation, Enterprise and Operational Risk, Compliance, Data Governance, Data Protection Officers (DPOs),Internal Audit, and IT/Technology Audit.
- Target Sectors: Written primarily for financial institutions—banks, Islamic banks, insurers, takaful operators,digital banks, e-money issuers, asset managers, and development financial institutions—operating under BNM and Securities Commission expectations. It is equally relevant to government-linked companies (GLCs),telecommunications, healthcare, utilities, organisations pursuing ISO/IEC 42001 certification, and assurance/advisory practitioners.
- Prerequisites: Participants should be comfortable reading reports and metrics. No coding or data science background is required
Practical Toolkit & Outputs: Participants work in groups on a common case organisation featuring three realistic AI models (an in-house credit decisioning model, a vendor-supplied fraud model, and a generative AI customer assistant) to produce six key deliverables to take back to their organisations:
1. An AI inventory with materiality and risk tiering scored on impact, complexity, and autonomy.
2. A model validation review checklist and a completed validation findings memo.
3. A fairness and explainability test plan with interpreted results, plus a human review protocol aligned to PDPC guidelines.
4. A generative AI/AI agent assurance test plan and a third-party AI due diligence sheet covering performance guarantees, data protection, change notification, audit rights, and concentration risk.
5. A monitoring specification setting drift and performance thresholds, triggers, escalation, and incident reporting.
6. A risk-based AI audit work programme mapped to BNM AI principles and ISO/IEC 42001, a one-page board assurance finding, and a customised 90-day assurance action plan
7. No Coding Required: All validation dashboards, fairness/explainability outputs, and generative AI evaluation results are prepared in advance by the trainer, allowing participants to focus entirely on reading, interpreting, and challenging evidence.
The course is structured across 5 main modules:
- Module 1: AI Model Risk Foundations: Lines of Defence and Regulatory Expectations Applies the three lines of defence to AI and details how machine learning, generative AI, and AI agents alter model risk through data dependence, drift, opacity, non-deterministic outputs, and autonomous actions. Sets out regulatory expectations from BNM, PDPC guidelines, the proposed AI Governance Bill, the AICB CRO Forum Framework, MAS, ISO/IEC 42001/23894, and AI Verify. Includes a hands-on activity building an AI inventory and tiering systems by materiality.
- Module 2: Validating AI Models: Conceptual Soundness, Data and Performance Focuses on establishing conceptual soundness, fitness for purpose, data lineage, data quality, and performance metrics beyond accuracy (precision, recall, calibration, error cost), alongside back-testing, champion vs. challenger comparisons, stability testing, and stress scenarios. Participants evaluate a prepared validation report for the credit model and write a validation findings memo.
- Module 3: Fairness, Explainability and Human Review of Automated Decisions Introduces fairness definitions/conflicts, proxy variable discrimination, and explainability outputs (SHAP values, feature importance,counterfactuals) using the Veritas FEAT methodology and PDPC guidance on human review. Participants interpret outputs and design a human review protocol for automated decisions.
- Module 4: Assuring Generative AI, AI Agents and Third-Party AI Covers evaluation sets,hallucination/groundedness measurement, red-teaming, prompt injection, guardrails, logging, human oversight,and autonomy tiers. Extends into vendor/third-party AI due diligence, contractual protections (performance guarantees, data protection, change notification, audit rights), and concentration risk.
- Module 5: Monitoring, Audit and Assurance Reporting Covers ongoing monitoring of data and concept drift,performance decay, setting thresholds/triggers, retraining governance, incident/near-miss reporting, and creating a risk-based AI audit work programme mapped to BNM AI principles and ISO/IEC 42001. Closes with writing executive findings for risk committees, boards, and regulators, and presenting 90-day action plans.
- Lead Trainer: Ir. Dr. Manjeevan Singh Seera
  - Qualifications: PhD in Computational Intelligence from Universiti Sains Malaysia.
  - Role & Affiliation: Associate Professor at the School of Business, Monash University Malaysia, and a registered Professional Engineer.
  - Accreditation: HRDCorp Registered Trainer (Trainer No. 16092).
  - Expertise: Specialises in machine learning, predictive analytics, AI governance, intelligent systems, and responsible AI. He prepares all pre-calculated dashboards and case materials, facilitates all group activities, and delivers the full two-day curriculum.
- Guest Speakers: A guest roundtable on Day 2 may feature a Head of Model Validation, Chief Risk Officer,Head of Internal Audit, Data Protection Officer, or Regulator (subject to guest availability).

Education and Training Academy

Education and Training Academy

AI Model RIsk, Validation and Assurance

AI Model Risk, Validation and Assurance