Ten modules. One competency: judgment over machines.

A plain-language account of what a large language model actually is, written for a numerate professional who has never seen a line of code and must never be assumed to have. It explains tokens, training and prediction; why generation is not retrieval; why hallucination is a structural property rather than a bug; and it draws the distinctions the rest of the course depends on — between narrow, generative, multimodal and agentic AI. By the end you can judge, from the mechanism rather than the marketing, where a tool can and cannot be trusted with money and personal data.

The evidence module. It walks you through how generative and agentic AI have rewritten the economics of financial crime: deepfake voice and video fraud, AI-synthesised identity documents, fraud-as-a-service, and multi-step, agent-driven schemes — anchored in the numbers and the named cases, including the South African picture. You leave able to describe the principal attack types, cite the leading evidence to make the internal case for AI-risk competency, and map your own institution's exposure surface.

The first of the platform's signature hands-on modules — and one nothing else in the South African market offers. It explains how synthetic identities, forged documents and deepfake audio and video are generated, then drills the detection of them: presentation and liveness attacks, artefacts in synthetic media, and the hardening of a know-your-customer and onboarding workflow against them. You leave with a repeatable detection routine.

How machine learning actually does transaction monitoring, where it fails, and how criminals now use automation against it. The module covers supervised and unsupervised detection, graph and network analytics for structuring, the false-positive problem, and the emerging threat of agent-driven laundering — all grounded in the South African FICA regime and the Risk Management and Compliance Programme obligation. You leave able to read, challenge and supervise a model-driven alert, and to articulate the human role that FATF expects to remain.

How bias actually enters automated credit decisions — through data, labels and proxies rather than intent — and how to find it, measure it and explain an adverse decision lawfully. The module teaches the principal fairness criteria and why they cannot all be satisfied at once, the mechanics of proxy discrimination and disparate impact, a structured fairness-review routine, and the legal frame: the National Credit Act's affordability regime, POPIA section 71's automated-decision rights, and the EU AI Act's classification of credit scoring as high-risk. It is written for a South African lending environment, where geography carries history.

The discipline of not trusting a model you cannot interrogate. It introduces model-risk management in the lineage of supervisory guidance, the main explainability methods and their genuine limits, and the difference between an explanation that satisfies a regulator and one that satisfies an engineer — framed for a South African institution and its Prudential Authority expectations. It closes the Core certificate by giving every prior module its governance spine: the verification routine, the monitoring stack and the fairness audit all assume a professional who can read, stress and challenge a model's documentation. This module builds that professional.

How to get the productivity of generative AI without leaking the personal information that regulation protects. The module shows how information actually escapes — through prompts, outputs, memorisation and training — then teaches the controls: the consumer-versus-enterprise distinction, safe prompting and redaction discipline, and the full POPIA mapping across security safeguards, the operator relationship, automated decisions and cross-border transfers. It is the module that lets an adviser safely use AI for a client report.

The security module. It covers how models are attacked rather than merely misused — adversarial evasion, training-data poisoning, and prompt injection against AI-integrated applications — and connects each to the concrete financial threat of manipulated inputs against fraud and underwriting models. It closes the loop with Module 5 by showing how a manipulated application defeats an automated decision, and it sets a realistic defensive posture: what can be hardened, what cannot, and where a human must remain in the loop.

What changes when AI stops answering and starts acting. The module explains agentic systems and tool use, then applies them to the advisory frontier: robo-advice, hyper-personalised recommendations, suitability and mis-selling risk, and the competitive position of the human adviser — alongside the supervision duties that agentic delegation creates. Delegation without abdication is the theme, and the FAIS conduct regime is the frame.

The synthesis module, culminating in an applied capstone. It maps the governance landscape — the EU AI Act, the NIST AI Risk Management Framework, ISO/IEC 42001 and 23894, the OECD Recommendation — and the financial-sector guidance of the FSB, BIS, IOSCO and the South African regulators, then guides the learner to assemble a real institutional AI-risk protocol: scope, permitted and forbidden uses, model governance, data rules, incident response and the three lines of defence. The capstone deliverable is that protocol, built from the artefacts every prior module produced and assessed against the whole course. This is where nine modules of competencies become one governable position.