AI Ethics and Governance Training Course

Participants can define governance roles, decision rights and reporting arrangements for AI.

• Board accountabilities under ISO/IEC 42001 AI management system requirements
• Three-lines model assigns ownership, challenge and independent assurance
• AI governance charter defines decision rights, escalation paths and reporting
• RACI matrix allocates model-owner, risk, legal and technology responsibilities
• Risk appetite statements set prohibited, restricted and approved AI uses
• Executive dashboards track inventory coverage, incidents, exceptions and assurance findings

Participants can identify applicable AI obligations and convert them into policy requirements.

• EU AI Act risk categories and operator obligations
• GDPR automated-decision provisions, lawful basis and data-subject rights
• Cross-border deployment considers jurisdiction, sector rules and contractual commitments
• ISO/IEC 23894 guidance applied to AI risk management
• NIST AI RMF functions: Govern, Map, Measure and Manage
• Policy hierarchy distinguishes mandatory controls, procedures, guidance and acceptable-use rules

Participants can classify AI systems and document their risks, impacts and treatment actions.

• AI system inventory captures purpose, owner, data, model and dependencies
• Use-case triage assesses affected people, decision materiality and potential harm
• Algorithmic impact assessment scopes rights, safety, equity and environmental effects
• Risk scenarios connect threat sources, failure modes, impacts and controls
• Likelihood-impact scoring calibrated with control effectiveness and residual-risk ratings
• Risk registers record treatment owners, due dates, evidence and acceptance

Participants can specify data and security controls appropriate to AI system risks.

• Data lineage maps collection, transformation, access, retention and deletion points
• Purpose limitation and minimisation tested against model feature requirements
• Data quality dimensions assess accuracy, completeness, timeliness, consistency and representativeness
• Consent, lawful basis and sensitive-data handling documented for each dataset
• Privacy impact assessment identifies mitigations for personal-data processing risks
• Secure development controls address poisoning, leakage, inversion and extraction attacks

Participants can evaluate fairness evidence and establish meaningful explanation and challenge processes.

• Fairness definitions compared: demographic parity, equalised odds and calibration
• Representative sampling tests subgroup coverage before training and deployment
• Disparate impact analysis quantifies outcome gaps across protected groups
• Confusion matrices reveal subgroup differences in false positives and negatives
• Explainability methods distinguish global model behaviour from individual explanations
• SHAP and counterfactual explanations support review, challenge and communication

Participants can establish evidence-based controls from AI design through retirement.

• Lifecycle gates define evidence required before build, release and retirement
• Model cards document intended use, limitations, evaluation and monitoring requirements
• Datasheets for Datasets record provenance, composition, collection and recommended use
• Validation plans specify acceptance thresholds, test sets and independent review
• Change-control procedures trigger revalidation after data, model or purpose changes
• Production monitoring detects drift, performance decay, bias and abnormal usage

Participants can assess and control material risks associated with generative AI systems.

• Generative AI use cases classified by autonomy, audience, sensitivity and consequence
• Prompt-injection, jailbreak and retrieval risks assessed using threat modelling
• Retrieval-augmented generation controls cover source quality, access and citation
• Human review thresholds set for customer, employee and public-facing outputs
• Content provenance, disclosure and synthetic-media labelling reduce deception risks
• Acceptable-use rules address confidential inputs, copyright and shadow AI

Participants can evaluate suppliers and design effective human oversight for AI-supported decisions.

• Vendor due diligence examines training data, testing, security and governance evidence
• Contract clauses specify audit rights, incident notice, data use and liability
• Service-level metrics cover accuracy, uptime, response times and remediation
• Human-in-the-loop designs assign authority, competence, timing and override mechanisms
• Meaningful human review prevents rubber-stamping of automated recommendations
• Procurement gates reject suppliers unable to evidence required controls

Participants can prepare AI incident processes and prioritise a governance implementation plan.

• AI incident taxonomy distinguishes harm, near miss, control failure and complaint
• Incident response playbooks assign containment, investigation, notification and recovery actions
• Root-cause analysis examines data, model, process, people and supplier factors
• Internal audit tests control design, operating evidence and exception handling
• Assurance reports communicate residual risk, findings, management actions and deadlines
• Ninety-day governance plan prioritises owners, milestones, measures and executive decisions