Domain 2: AI Governance, Ethics, Risk, and Compliance (17% of exam)

EU AI Act

The most comprehensive binding AI regulation referenced on the exam, which classifies AI systems by risk tier (unacceptable, high, limited, minimal) and imposes obligations scaled to that risk level.

NIST AI RMF

Risk Management Framework — a voluntary US framework organized around four core functions: Govern (culture and policy), Map (context and risk identification), Measure (analyze and assess risk), and Manage (respond and prioritize).

ISO/IEC 42001

The international AI management system standard, following the same structural approach as ISO 27001 (ISMS) but scoped to how an organization governs AI responsibly across its lifecycle.

GDPR

General Data Protection Regulation — EU data protection law that applies directly to AI systems processing personal data during training or inference, enforcing principles such as lawful basis, data minimization, and the right to erasure.

AI governance vs. traditional IT governance

Traditional IT governance assumes deterministic systems with stable behavior (focusing on uptime and change control), whereas AI governance must also account for non-determinism, drift, emergent behavior, and model-specific risk.

AI use-case inventory

A centralized, maintained list of every AI system and use case in the organization so leadership knows what AI exists and can govern it (preventing unmonitored shadow AI).

Organizational AI roles

Key governance stakeholders including the Chief AI Officer (executive accountable for AI strategy and risk), AI Ethics Officer (oversees responsible AI principles), and AI Champions (embedded advocates and points of contact within business units).

Core Responsible AI principles

Six foundational ethical principles for AI systems: fairness, reliability, safety, privacy, inclusiveness, and transparency.

Fairness (Responsible AI)

Ensuring that AI systems treat all individuals and demographic groups equitably without producing systemic, unjustified, or discriminatory bias.

Reliability (Responsible AI)

The ability of an AI system to perform its intended functions consistently, accurately, and predictably under normal and stressed operating conditions.

Safety (Responsible AI)

Designing and operating AI systems to prevent physical, digital, psychological, or operational harm to humans, infrastructure, and society.

Privacy (Responsible AI)

Protecting personal, proprietary, and sensitive data throughout the entire AI lifecycle, from data collection and ingestion to storage and inference.

Inclusiveness (Responsible AI)

Designing AI solutions that empower and serve a diverse population, accommodating different physical abilities, languages, cultures, and socioeconomic backgrounds.

Transparency (Responsible AI)

Making the operation, data sources, decision paths, and limitations of an AI system clear and understandable to users and stakeholders.

Algorithmic bias mitigation

Actively testing and adjusting an AI model so it does not produce systematically unfair outcomes across protected groups such as race, gender, or age.

Transparency vs. explainability

Transparency means disclosing that AI is being used and generally how it works, whereas explainability means being able to articulate why a specific model made a specific decision.

Systemic impacts and sustainability

Considering broader societal effects of AI deployment—such as energy and compute consumption, job displacement, and concentration of power—beyond the immediate technical use case.

AI risk quantification and mapping

Translating AI-specific risks (such as bias, drift, hallucination, and autonomous action) into measurable, comparable terms within a risk register, applying traditional risk discipline to AI.

AI impact assessments

Evaluating potential harm from an AI system, including its blast radius (how far-reaching a failure could be) and the elevated inherent risk of autonomous actions taken without human approval.

Use-case risk tiers

Categorizing AI systems by combining human impact (potential harm) and technical complexity so higher-impact, higher-complexity use cases receive stricter controls, echoing the EU AI Act's risk-tier logic.

AI risk treatments

Applying the standard risk management options—avoid, mitigate, transfer, and accept—specifically to AI system components.

Reassessment triggers (AI risk)

Events that force a fresh AI risk review—such as model updates, detected behavioral drift, or new vulnerability disclosures—reflecting that AI risk assessment is continuous rather than one-and-done.

Regulated and sensitive data types

Data categories such as PII (personally identifiable information), PHI (protected health information), financial data, and export-controlled data

IP, copyright, and training data fair use

Legal risk surrounding whether AI training data was lawfully obtained or licensed, representing a live, unsettled legal area where organizations face significant exposure.

Data derivation / regulatory contamination

The risk that a model trained on regulated data absorbs that regulated status into its outputs—for example, a model trained on PHI leaking PHI-like patterns even without directly storing records.

Data retention and deletion management (AI)

Defining lifecycles for how long AI training and inference data is kept and properly deleted, complicated by the fact that a trained model may retain or remember patterns from data intended for deletion.

Consent mechanisms (AI)

Ensuring that data used for AI training or inference—including publicly available data—was collected with appropriate user consent or a valid legal basis.

Third-party AI due diligence

Vetting external AI vendors' security controls and posture before adopting their models or platforms.

Procurement controls, security SLAs, and incident notification rights

Contract-level protections for third-party AI services, such as requiring a vendor to notify the organization within a specified number of hours of a breach affecting its data or models.

AIBOM

AI Bill of Materials — an inventory of AI component dependencies (data, models, libraries) that provides visibility into what is inside an AI system not built from scratch.

Shared responsibility models (AI)

Clarifying which security controls the AI vendor is responsible for versus what the customer must secure, extending cloud shared-responsibility logic to AI services.

Data center security / Trusted Execution Environments (TEEs) / confidential computing

Hardware-level data center protections ensuring data stays encrypted and isolated even while being actively processed, protecting AI workloads running on infrastructure the organization does not fully control.

HITL vs. HOTL

Human-in-the-Loop (HITL) requires a human to actively review and approve each action (gatekeeping), whereas Human-on-the-Loop (HOTL) allows greater model autonomy where a human monitors and can intervene without approving every action (supervisory oversight).

Intervention, override, and sign-off authorities

Clearly defined organizational roles specifying who has the authority to stop, reverse, or approve an AI system's actions.

Shadow AI detection

Finding unauthorized or unsanctioned AI tools or models in use within the organization that governance is unaware of, directly supporting the completeness of the AI use-case inventory.

Governance artifacts

The documentary paper trail—such as model documentation, decision logs, and exception records—that proves AI governance is actively practiced rather than merely declared on paper.

AI interaction labeling

Disclosing when content or an interaction is synthetic or AI-generated (such as deepfake disclosure requirements) so humans know they are viewing or interacting with AI output.

Internal and external AI audits

Periodic formal reviews of AI systems to verify compliance, check for bias and drift, and confirm whether the systems are delivering their claimed business outcomes.

Skills assessments and gap analyses (AI)

Evaluating whether staff operating, approving, or securing AI systems (such as data scientists, approvers, and security personnel) possess the competence required for their roles.

Role-based AI training

Tailoring AI security and ethics education to specific job functions, recognizing that developers need different training than executives who approve AI use cases.

Training effectiveness metrics

Measuring whether AI security training actually changed behavior and reduced risk—rather than just tracking attendance—to foster a risk-aware organizational culture.