Courseiva

ISACA Advanced in AI Risk (AAIR) (AAIR) (AAIR) — Questions 1–75

206 questions total · 3 pages · All types, answers revealed

Take a mock examExam hub

Page 1 of 3

Page 2
1
Multi-Selectmedium

Which TWO actions should be taken when a model exhibits significant drift?

Select 2 answers
A.Perform a root cause analysis to determine the nature of the drift.
B.Retrain the model on updated data if the drift is identified as environmental change.
C.Disable all logging to improve performance.
D.Change the model architecture to something simpler.
E.Immediately increase the compute resources.
AnswersA, B

Drift can stem from data quality issues, schema changes, or genuine environmental shifts, each demanding a different remedy. Root cause analysis identifies which mechanism is responsible before remediation, preventing unnecessary retraining and ensuring the subsequent action targets the actual cause of the degradation.

Why this answer

Option A is correct because when a model exhibits significant drift, the first step is to perform a root cause analysis to determine whether the drift is due to data quality issues, concept drift, or environmental change, which informs the appropriate remediation. Option B is correct because if the root cause analysis identifies the drift as environmental change (e.g., shifting user behavior or seasonal patterns), retraining the model on updated, representative data is the standard remedy to restore performance. Option C is incorrect because disabling logging removes the observability needed to detect and diagnose drift, and it does not address the underlying model degradation.

Option D is incorrect because switching to a simpler architecture is not a targeted response to drift and may reduce performance without resolving the data or concept shift. Option E is incorrect because adding compute resources does not fix drift, since the issue is a mismatch between the model and current data distribution, not insufficient processing power.

Exam trap

The trap is choosing operational quick fixes (more compute, simpler model) instead of diagnostic and data-centric actions; the exam tests understanding that drift is a data problem, not an infrastructure one.

Full explanation →
2
MCQmedium

You are the AI risk lead at a mid-sized insurer. The Chief Risk Officer asks you to establish a formal AI risk taxonomy that will be used consistently across all business units for identifying, assessing, and reporting AI risks. Which action best ensures the taxonomy will be adopted and remain useful over time?

A.Delegate taxonomy design entirely to the data science team, since they understand AI models best and can define the risks most accurately.
B.Map the taxonomy to the enterprise risk management framework and existing risk register categories, then pilot it with two business units before enterprise rollout.
C.Draft the taxonomy with only legal and compliance input, then publish it as a mandatory standard to all business units to ensure regulatory alignment.
D.Adopt the risk categories directly from a leading industry publication without modification, since external benchmarks carry more credibility than internal design.
AnswerB

Aligning the new AI risk taxonomy to the existing ERM framework and risk register categories ensures that AI risks are not treated as a separate silo, enabling consistent aggregation and reporting. Piloting with two business units surfaces practical gaps and builds internal advocates before full rollout, increasing adoption and long-term usefulness.

Why this answer

A durable AI risk taxonomy must integrate with existing ERM structures so risks can be aggregated and compared across the enterprise. Piloting with a subset of business units validates the categories against real AI use cases and generates feedback before broad rollout. This combination of alignment and iterative validation is what drives sustained adoption and keeps the taxonomy relevant as AI use evolves.

Exam trap

The trap here is assuming that adopting an authoritative external taxonomy or delegating design to a single expert team guarantees quality, when adoption actually depends on integration with existing ERM structures and input from multiple stakeholders.

Full explanation →
3
MCQhard

You are mapping AI risks to the NIST AI RMF. Which step is essential when documenting the 'Map' function for a high-risk autonomous system?

A.Identifying and documenting the context and intended use
B.Updating the SOC2 Type II report
C.Automating the model deployment pipeline
D.Conducting a quarterly penetration test
AnswerA

The Map function establishes context: intended purpose, users, affected populations, operating environment and constraints. Documenting these for a high-risk autonomous system frames all subsequent risk identification and measurement, ensuring later Assess and Manage activities target the actual deployment scenario.

Why this answer

Identifying and documenting the context and intended use is the foundation of mapping AI risks to an enterprise framework.

Full explanation →
4
Multi-Selectmedium

An AI risk practitioner is designing the human oversight component for a generative AI assistant used by customer service agents. The assistant drafts responses that agents may send after review. Which TWO controls most directly reduce the risk of harmful or non-compliant outputs reaching customers? (Choose two.)

Select 2 answers
A.Increasing the model's temperature setting to produce more varied and natural-sounding responses.
B.Running automated content filters that block outputs containing prohibited topics or personally identifiable information.
C.Requiring agents to affirmatively edit or approve each draft before it can be sent, with the approval logged.
D.Expanding the assistant's knowledge base with additional product documentation to improve answer coverage.
E.Reducing the number of agents authorised to use the assistant to a small pilot group.
AnswersB, C

Automated filters act as a deterministic guardrail that catches prohibited content or PII before any human sees or sends it. They complement human review by reducing the volume and severity of dangerous drafts. Together with logged approval, filters form a defence-in-depth control set for generative output risk.

Why this answer

Effective oversight for a generative assistant combines a human decision point with an automated safety layer. Requiring logged affirmative approval ensures no draft reaches a customer without human accountability, while content filters catch prohibited or sensitive material deterministically. Together they reduce both the likelihood and the impact of harmful outputs, which is the objective of human oversight in the AI lifecycle.

Exam trap

The trap here is treating model tuning or rollout scoping as oversight controls, when oversight specifically requires a human decision point plus a deterministic output guardrail.

Full explanation →
5
Multi-Selecthard

The AI Governance Committee is defining the roles for AI Risk oversight. Which TWO functions are essential for the 'Three Lines of Defense' model in AI Risk?

Select 2 answers
A.AI development teams performing self-assessments.
B.The AI Ethics Committee drafting external press releases.
C.The Chief Information Officer managing all AI infrastructure.
D.External vendors providing cloud infrastructure.
E.The AI Risk Management function providing independent oversight.
AnswersA, E

In the Three Lines of Defense model, the first line comprises operational and development teams who own and manage risk directly, performing self-assessments of their own AI systems. This makes AI development teams' self-assessments an essential first-line function.

Why this answer

In the Three Lines of Defense model, the first line consists of the operational teams that own and manage risk directly, so option A (AI development teams performing self-assessments) is correct because these teams are the ones building and running AI systems and must perform self-assessments as their first-line risk ownership responsibility. The second line provides independent oversight, monitoring, and challenge to the first line, so option E (the AI Risk Management function providing independent oversight) is correct because it is the dedicated risk function that establishes policies, frameworks, and independent review over AI risks. Option B is incorrect because drafting external press releases is a communications activity, not a risk oversight function within the Three Lines of Defense.

Option C is incorrect because managing AI infrastructure is an operational/technology responsibility that would sit in the first line, not a distinct defense layer for AI risk oversight. Option D is incorrect because external vendors supplying cloud infrastructure are third parties, not an internal line of defense in the model.

Exam trap

AAIR often tests the tendency to confuse operational or communications roles with the independent oversight functions required by the Three Lines of Defense model, so candidates must distinguish first-line ownership from second-line oversight.

Full explanation →
6
MCQhard

A firm is using the NIST AI Risk Management Framework to document its AI risk program. Which approach best ensures that AI risk metrics are dynamic?

A.Conducting annual risk assessments
B.Setting static thresholds for model accuracy
C.Limiting AI deployment to offline systems
D.Integrating model performance metrics into operational dashboards
AnswerD

Feeding model performance metrics into operational dashboards keeps AI risk measurement continuous rather than point-in-time, satisfying the stem's demand for dynamic metrics. Live telemetry surfaces drift, degradation and emerging harms as they occur, enabling the Govern and Measure functions of the NIST AI Risk Management Framework to trigger timely response.

Why this answer

Integrating model performance metrics into operational dashboards ensures that AI risk metrics are dynamic, as they are continuously updated and monitored in real time. This allows for timely detection of drift, bias, or degradation. Annual assessments and static thresholds are not dynamic.

Exam trap

AAIR often tests the concept of dynamic risk metrics, and candidates may incorrectly choose periodic assessments or static thresholds, confusing thoroughness with dynamism.

How to eliminate wrong answers

Option A is wrong because annual risk assessments are periodic and static, not dynamic; they do not provide continuous monitoring. Option B is wrong because setting static thresholds for model accuracy is a fixed, non-adaptive approach that does not reflect changing conditions. Option C is wrong because limiting AI deployment to offline systems does not make risk metrics dynamic; it merely restricts deployment scope.

Full explanation →
7
Multi-Selectmedium

Which TWO documents are essential to include in the AI Risk Project 'Review Pack'?

Select 2 answers
A.A summary of the team's lunch schedule.
B.A detailed risk mitigation and action plan.
C.A list of all software installed on company laptops.
D.A copy of the company's lease agreement.
E.A comprehensive AI Risk Assessment report.
AnswersB, E

A detailed risk mitigation and action plan is essential because the Review Pack must evidence how identified AI risks will be treated, not merely list them. It satisfies the stem's requirement for documents supporting governance review, showing owners, controls and timelines that demonstrate active risk management to approvers.

Why this answer

Option B, a detailed risk mitigation and action plan, is essential because the Review Pack must show how identified AI risks will be treated, including specific controls, owners, timelines, and residual-risk decisions that reviewers can evaluate. Option E, a comprehensive AI Risk Assessment report, is essential because it provides the core evidence of risk identification, analysis, and evaluation, such as model purpose, data sources, affected stakeholders, likelihood/impact ratings, and inherent risk levels. Together, B and E give the Review Pack both the assessment basis and the remediation roadmap needed for governance sign-off.

The unmarked options do not belong: A (team lunch schedule) and D (company lease agreement) are unrelated administrative documents, and C (list of all software installed on company laptops) is an IT asset inventory that, while potentially useful elsewhere, is not one of the two essential AI risk review documents in this scenario.

Exam trap

AAIR often tests whether candidates can distinguish governance/risk artifacts from generic IT or administrative documents, so the trap is selecting plausible-sounding but irrelevant operational documents instead of the two core risk deliverables.

Full explanation →
8
MCQeasy

What is the first step in performing an AI Risk Assessment?

A.Hiring a security consultant.
B.Updating the company website.
C.Running a bias test.
D.Defining the scope, context, and purpose of the AI system.
AnswerD

Defining scope, context and purpose establishes the boundaries, intended use and environment of the AI system, which every subsequent risk identification, analysis and evaluation activity depends on. Without this foundation, risks cannot be assessed against the correct operational and regulatory criteria.

Why this answer

Per ISO 31000 and NIST AI RMF, every risk assessment begins with establishing scope, context, and purpose — defining which AI system is being assessed, its boundaries, stakeholders, operating environment, and the objectives of the assessment. Without this foundational step, subsequent activities like bias testing or control selection lack a reference frame and may address the wrong risks or miss critical ones. This mirrors ISO 31000 Clause 6.3.1, which explicitly places 'defining scope and context' as the first activity in the risk assessment process.

Exam trap

AAIR often tests whether candidates recognize that scoping/context-setting is the mandatory first step, versus jumping to technical activities like bias testing that presuppose scope is already defined.

How to eliminate wrong answers

Option A is wrong because hiring a security consultant is a resourcing decision, not a methodological step — consultants may assist with the assessment but their engagement does not constitute the first analytical step. Option B is wrong because updating the company website is entirely unrelated to risk assessment methodology and represents a distractor. Option C is wrong because running a bias test is a specific technical assessment activity that presupposes you already know which model, dataset, and fairness dimensions are in scope — it cannot logically precede scoping.

Full explanation →
9
Multi-Selecthard

You are integrating AI governance into the existing Corporate Governance framework. Which THREE of the following activities are essential to ensure the Board of Directors maintains adequate oversight?

Select 3 answers
A.Reviewing and approving the organization's AI Risk Appetite Statement.
B.Installing software patches on all GPU servers.
C.Designating executive-level accountability for AI risk management outcomes.
D.Conducting daily code reviews of all AI training algorithms.
E.Establishing a reporting cadence for significant AI-related risk incidents and model performance trends.
AnswersA, C, E

Reviewing and approving the AI Risk Appetite Statement is a core Board responsibility, since setting risk tolerance is a governance function that cannot be delegated to management. This activity satisfies the stem's oversight constraint by ensuring the Board formally defines how much AI risk the organization is willing to accept before any AI initiative proceeds.

Why this answer

Option A is correct because the Board's core governance role includes setting and approving the organization's risk appetite, and an AI Risk Appetite Statement translates that mandate into explicit tolerance thresholds for AI-specific risks (e.g., bias, model drift, data privacy), giving the Board a formal basis for oversight. Option C is correct because designating executive-level accountability (e.g., a Chief AI Officer or an AI risk owner at the C-suite) establishes clear ownership and escalation paths, which is a fundamental governance control ensuring the Board can hold management responsible for AI risk outcomes. Option E is correct because a defined reporting cadence for significant AI-related risk incidents and model performance trends provides the Board with timely, recurring visibility into AI risk posture, enabling informed oversight and intervention consistent with corporate governance principles.

Options B and D are not Board-level governance activities: patching GPU servers is an operational IT maintenance task, and daily code reviews of training algorithms are tactical engineering controls, both of which belong to management and technical teams rather than the Board's oversight responsibilities.

Exam trap

AAIR often tests the confusion between operational/technical tasks (patching, code review) and genuine Board-level governance activities (appetite approval, accountability designation, reporting cadence).

Full explanation →
10
Multi-Selectmedium

Which TWO factors should be included in an AI risk appetite statement?

Select 2 answers
A.The intended impact on company revenue
B.The specific programming language allowed
C.The preferred office software
D.Financial thresholds for model-related incidents
E.Maximum tolerable level of model bias
AnswersD, E

Quantified financial thresholds translate AI risk appetite into measurable tolerance for model-related incidents, such as maximum acceptable loss per event. This gives governance bodies an objective limit for escalation and remediation decisions rather than vague qualitative statements.

Why this answer

Option D is correct because a risk appetite statement must quantify how much financial loss from model-related incidents (e.g., fraud, erroneous decisions, remediation costs) the organization is willing to tolerate, expressed as concrete monetary thresholds or limits. Option E is correct because AI-specific risk appetite must define the maximum acceptable level of model bias (e.g., disparate impact ratios or fairness metric tolerances), since bias is a core AI risk affecting ethics, compliance, and legal exposure. Options A, B, and C are not appropriate: intended revenue impact is a business outcome rather than a risk tolerance limit, programming language is an implementation/technical standard, and preferred office software is an unrelated productivity choice — none of these express acceptable risk levels for AI systems.

Exam trap

AAIR often tests the confusion between business objectives (revenue impact) and risk tolerance metrics (financial thresholds, bias limits), leading candidates to select revenue as a risk appetite factor.

Full explanation →
11
MCQhard

A developer is using 'Shadow Deployments' for a new AI model. What is the main risk being addressed?

A.Operational risk of deploying a poorly performing model to production.
B.Model documentation compliance.
C.Training data leakage.
D.Hardware infrastructure costs.
AnswerA

Shadow deployment mirrors live traffic to the new model without serving its outputs, letting teams compare performance against the incumbent and catch degradation before promotion, directly addressing the operational risk of releasing a poorly performing model.

Why this answer

Shadow deployments allow real-world data validation without exposing end-users to potential model failures.

Full explanation →
12
MCQmedium

A firm adopts the 'Three Lines of Defense' model for AI. What is the specific responsibility of the 'Third Line' (Internal Audit)?

A.Providing independent assurance on the design and effectiveness of the AI governance framework.
B.Approving AI model deployment.
C.Developing the AI models.
D.Defining the AI risk appetite.
AnswerA

Internal Audit sits outside management, so it provides independent assurance that the AI governance framework is designed and operating effectively. This satisfies the stem's Third Line definition, distinct from the First Line's ownership and the Second Line's oversight.

Why this answer

The Third Line of Defense in the Three Lines of Defense model is Internal Audit, which provides independent, objective assurance to the board and senior management on the design and operating effectiveness of the AI governance framework, including risk management and controls. It does not own or operate AI controls; it evaluates them. This independence is what distinguishes it from the First Line (business/IT operations) and Second Line (risk/compliance).

Exam trap

The trap here is confusing the oversight role of the Second Line (risk/compliance) with the independent assurance role of the Third Line (Internal Audit) — candidates often pick 'approving' or 'defining' activities that actually belong to management or the board.

How to eliminate wrong answers

Option B is wrong because approving AI model deployment is a First Line (business/technology owner) or governance committee responsibility, not an independent audit function — auditors must not approve the very things they later audit. Option C is wrong because developing AI models is a First Line operational activity performed by data science/engineering teams, which would destroy audit independence. Option D is wrong because defining the AI risk appetite is a board and executive management responsibility, often informed by the Second Line (risk function), not Internal Audit.

Full explanation →
13
MCQhard

When reporting AI risk to the board, which approach best demonstrates 'Risk Culture' maturity?

A.Listing all AI projects currently in production with their budget spend.
B.Reporting solely on the number of successful cyber attacks against the AI infrastructure.
C.Presenting a heatmap showing residual risk levels against the defined enterprise risk appetite.
D.Providing a list of all AI models currently undergoing training.
AnswerC

A residual-risk heatmap mapped to the enterprise risk appetite evidences mature risk culture because it shows governance actively comparing tolerated exposure against agreed limits, not merely listing threats. This satisfies the stem's reporting constraint: the board sees quantified, appetite-aligned decisions, demonstrating that risk ownership and escalation are embedded rather than ad hoc.

Why this answer

Presenting a heatmap of residual risk against the enterprise risk appetite demonstrates Risk Culture maturity because it shows the board not just what risks exist, but how they compare to the organization's stated tolerance thresholds. This reflects an integrated, governance-driven approach where risk is quantified, contextualized, and actively managed rather than merely inventoried. Mature risk culture requires translating technical AI risk into business-relevant terms that support informed decision-making at the board level.

Exam trap

The trap here is confusing activity reporting (projects, budgets, attacks) with risk reporting — candidates pick the option that sounds most 'executive' rather than the one that actually ties risk to appetite.

How to eliminate wrong answers

Option A is wrong because listing AI projects with budget spend is an asset/inventory report, not a risk report — it says nothing about likelihood, impact, or tolerance. Option B is wrong because counting successful cyber attacks is a lagging, narrow security metric that ignores model risk, bias, drift, and other AI-specific risks. Option D is wrong because listing models in training is an operational status update with no risk quantification or appetite alignment.

Full explanation →
14
MCQeasy

Which document is essential for defining the roles and responsibilities within an AI Risk Program?

A.AI Governance RACI Matrix.
B.Data Privacy Impact Assessment (DPIA).
C.Model Training Log.
D.AI Incident Response Plan.
AnswerA

The AI Governance RACI Matrix directly satisfies the stem's requirement to define roles and responsibilities, assigning Responsible, Accountable, Consulted and Informed parties across AI risk activities. Unlike broader policy or charter documents, it specifies who owns each task, making accountability explicit within the AI Risk Program.

Why this answer

The AI Governance RACI Matrix is the document that explicitly defines who is Responsible, Accountable, Consulted, and Informed for AI risk activities across the organization. It is the standard governance artifact for assigning roles and responsibilities, ensuring clear ownership and avoiding gaps or overlaps in the AI Risk Program.

Exam trap

AAIR often tests whether candidates can distinguish governance/roles documents from risk-assessment, technical-log, or incident-response artifacts, so the trap is selecting a document that mentions roles (like an IR Plan) instead of the one specifically designed to define program-wide responsibilities.

How to eliminate wrong answers

Option B is wrong because a Data Privacy Impact Assessment (DPIA) assesses privacy risks of a specific processing activity — it documents risk analysis, not role assignments. Option C is wrong because a Model Training Log records training runs, hyperparameters, and datasets — it is a technical reproducibility artifact, not a governance roles document. Option D is wrong because an AI Incident Response Plan defines how to detect, respond to, and recover from AI incidents; it assigns incident-specific roles but does not define the overall program's roles and responsibilities.

Full explanation →
15
MCQhard

When integrating AI risk into the broader ERM (Enterprise Risk Management) framework, which approach is most effective?

A.Only report AI risks if they result in a financial loss
B.Keep AI risks in a separate, isolated tracking system
C.Map AI risks to existing risk categories (e.g., operational, compliance)
D.Automate the removal of AI risks from the main risk register
AnswerC

Mapping AI risks onto established categories such as operational and compliance lets organisations reuse existing ERM governance, appetite statements, and reporting lines. This integrates AI into current processes rather than creating a parallel framework, satisfying the goal of embedding AI risk within broader enterprise risk management.

Why this answer

Mapping AI risks to existing risk categories such as operational, compliance, and strategic risks integrates AI into the broader ERM framework. This allows the organization to leverage established risk management processes, reporting lines, and mitigation strategies. It ensures AI risks are not siloed and receive appropriate visibility and resourcing.

Exam trap

The trap is thinking AI risks are so unique they must be managed separately, leading candidates to choose isolation over integration into ERM.

How to eliminate wrong answers

Option A is wrong because reporting AI risks only after financial loss is reactive and ignores non-financial impacts like reputational harm, regulatory penalties, and ethical violations. Option B is wrong because keeping AI risks in a separate, isolated system prevents holistic risk view and may lead to inconsistent treatment and duplication of efforts. Option D is wrong because automating removal of AI risks from the main risk register would eliminate oversight and is contrary to the goal of integration.

Full explanation →
16
Multi-Selecthard

Which THREE items should be included in an AI 'Risk Assessment' report for a new project?

Select 3 answers
A.Identification of potential failure modes and their impact.
B.The favorite color of the lead developer.
C.The residual risk level after applying controls.
D.A list of all competitors currently using the same software.
E.Documentation of mitigating controls for identified risks.
AnswersA, C, E

Identifying potential failure modes and their impact directly satisfies the risk assessment requirement by mapping what could go wrong and the severity of each outcome. This structured analysis lets stakeholders prioritise mitigations before deployment, covering both likelihood and consequence dimensions essential to a defensible AI risk report.

Why this answer

A complete report covers the risk identification, the impact analysis, and the control strategy.

Full explanation →
17
MCQeasy

Why is 'Explainability' considered a key risk management control in the AI lifecycle?

A.It guarantees the model will never fail.
B.It increases the inference speed of the model.
C.It enables identification of potential bias, errors, or unexpected behavior in model decision-making.
D.It allows the model to train on less data.
AnswerC

Explainability exposes the features and reasoning driving each model output, letting reviewers trace decisions back to input data and detect bias, errors, or unexpected behaviour before harm occurs. This directly satisfies the stem's risk management constraint by making model decision-making auditable and correctable throughout the AI lifecycle.

Why this answer

Explainability provides visibility into how a model arrives at its outputs, which allows teams to detect bias, errors, and unexpected behavior before they cause harm. In risk management terms, it is a detective and preventive control that supports incident investigation, regulatory compliance, and stakeholder trust. Without it, failures can go undiagnosed and accountability becomes impossible.

Exam trap

The trap is conflating explainability with performance or data-efficiency benefits — candidates may pick an option that sounds like an operational advantage instead of recognizing explainability as a governance and risk-detection control.

How to eliminate wrong answers

Option A is wrong because no control can guarantee a model will never fail — explainability reduces risk but does not eliminate failure modes. Option B is wrong because explainability techniques such as SHAP or LIME add computational overhead and typically slow inference rather than speed it up. Option D is wrong because explainability has no bearing on the volume of training data required; data efficiency comes from techniques like transfer learning or data augmentation, not interpretability.

Full explanation →
18
MCQmedium

What is the primary benefit of documenting 'AI Model Lineage'?

A.It ensures the model is smaller in file size.
B.It provides transparency, auditability, and the ability to reproduce models.
C.It helps the marketing team write better ads.
D.It makes the model run faster.
AnswerB

AI model lineage records the datasets, code, parameters and transformations that produced each model version. This traceable chain delivers transparency, supports audits, and lets teams reproduce or rebuild a model exactly, which is the primary benefit the question asks for.

Why this answer

Documenting AI Model Lineage provides transparency, auditability, and reproducibility by capturing the full chain of data sources, transformations, hyperparameters, code versions, and dependencies used to produce a model. This record allows auditors and regulators to trace how a model was built, verify compliance with policies, and reproduce or re-validate the model when needed. It is a foundational governance artifact rather than a performance optimization.

Exam trap

AAIR often tests the misconception that governance documentation like lineage improves model performance or efficiency, when its actual purpose is transparency, auditability, and reproducibility.

How to eliminate wrong answers

Option A is wrong because model lineage documentation has no effect on the file size of a model; file size is determined by architecture, parameter count, and quantization, not by documentation. Option C is wrong because lineage is a technical and governance record for traceability and compliance, not a marketing input for advertising copy. Option D is wrong because documentation does not change runtime performance; inference speed depends on model architecture, hardware, and optimization techniques such as quantization or pruning, not on lineage records.

Full explanation →
19
MCQmedium

When documenting an AI Risk Program, what should be included in the 'AI Asset Register'?

A.The raw hardware serial numbers of the servers.
B.The model purpose, data lineage, stakeholders, and risk classification.
C.The exact lines of code in the model.
D.The home addresses of all developers.
AnswerB

The AI Asset Register inventories each AI system, so it must capture model purpose, data lineage, stakeholders, and risk classification. These attributes let the organisation track ownership, provenance, and risk tiering across its AI estate.

Why this answer

An AI Asset Register is an inventory of AI systems and models that captures governance-relevant metadata: the model's purpose, the data sources and lineage used to train it, the stakeholders accountable for it, and its risk classification. This information enables risk management, regulatory reporting, and lifecycle oversight across the AI portfolio.

Exam trap

AAIR often tests whether candidates distinguish governance-level metadata from low-level technical or personal details; the trap is selecting code or hardware specifics instead of purpose, lineage, stakeholders, and risk classification.

How to eliminate wrong answers

Option A is wrong because raw hardware serial numbers are infrastructure asset management details, not AI governance metadata; they do not describe the model's purpose, risk, or accountability. Option C is wrong because exact lines of code are implementation artifacts managed in version control, not appropriate for an asset register, which should remain at a governance level. Option D is wrong because developers' home addresses are personal data with no relevance to AI asset governance and would themselves create a privacy risk.

Full explanation →
20
MCQeasy

In the context of AI risk management, what does the 'Human-in-the-Loop' (HITL) concept primarily aim to achieve?

A.To reduce the cost of model development.
B.To maintain accountability and oversight for AI-driven decisions.
C.To increase the speed of model inference.
D.To automate the labeling of training data.
AnswerB

Human-in-the-Loop inserts human review at decision points, ensuring a person remains accountable for AI outputs rather than delegating responsibility to the model. This oversight mechanism satisfies governance requirements by enabling intervention before consequential actions are taken.

Why this answer

Human-in-the-Loop (HITL) primarily ensures that a human reviews, approves, or can override AI-driven decisions, preserving accountability and meaningful oversight. It addresses the governance need that a responsible person remains answerable for outcomes, especially in high-risk or regulated contexts. This maps directly to accountability principles in the NIST AI RMF and OECD AI Principles.

Exam trap

AAIR often tests whether candidates understand HITL as a governance/accountability control rather than a performance or cost optimization, so the trap is selecting efficiency-oriented answers that invert HITL's actual trade-off (it adds human effort to preserve oversight).

How to eliminate wrong answers

Option A is wrong because reducing model development cost is a financial/engineering outcome, not the purpose of HITL — HITL often increases cost by adding human review. Option C is wrong because increasing inference speed is a performance goal; HITL typically slows the pipeline because a human must intervene. Option D is wrong because automating training-data labeling is a data-annotation workflow (often called human-in-the-loop labeling, but that is a narrow ML-ops use case, not the risk-management purpose of accountability and oversight).

Full explanation →
21
MCQmedium

A practitioner is reviewing the 'Model Card' for an AI system. What is the primary purpose of this artifact in the AI lifecycle risk management process?

A.To automate the retraining pipeline for the model.
B.To serve as the primary authentication token for API access.
C.To provide transparency regarding the model's limitations, intended use cases, and performance metrics.
D.To define the deployment strategy in Kubernetes.
AnswerC

A Model Card documents intended use cases, performance metrics and known limitations, giving reviewers the transparency needed to assess fitness and residual risk. This satisfies the stem's focus on lifecycle risk management by exposing where the model may fail or be misapplied.

Why this answer

The Model Card is a standardized transparency artifact (popularized by Mitchell et al., 2019) that documents a model's intended use, limitations, performance metrics across subgroups, training data provenance, and ethical considerations. In AI lifecycle risk management, it enables downstream consumers, auditors, and governance teams to assess whether a model is appropriate for a given context and to identify residual risks. It is a documentation and disclosure mechanism, not an operational or security component.

Exam trap

AAIR often tests whether candidates confuse documentation artifacts (Model Cards, datasheets) with operational artifacts (pipelines, tokens, manifests) — the trap is picking a functional-sounding option when the question asks about a governance/transparency purpose.

How to eliminate wrong answers

Option A is wrong because retraining pipelines are automated via MLOps tooling (e.g., SageMaker Pipelines, Kubeflow, Airflow), not via a Model Card, which is a static documentation artifact. Option B is wrong because API authentication is handled by tokens, API keys, OAuth credentials, or IAM roles — a Model Card has no cryptographic or authentication function. Option D is wrong because Kubernetes deployment strategies (rolling, blue/green, canary) are defined in manifests, Helm charts, or service mesh configs, not in a Model Card.

Full explanation →
22
Multi-Selectmedium

Which TWO stakeholders are most important to engage when establishing an AI risk appetite?

Select 2 answers
A.The building security personnel.
B.The Board of Directors or Executive Committee.
C.The office cleaning staff.
D.Legal and Compliance officers.
E.The company's social media followers.
AnswersB, D

The Board or Executive Committee owns organisational risk appetite and holds accountability for accepting residual AI risk. Their engagement satisfies the stem because appetite set without executive mandate lacks authority, budget and enforcement across business units.

Why this answer

The Board of Directors or Executive Committee (B) is essential because setting AI risk appetite is a governance responsibility — this body defines the organization's overall tolerance for risk, approves strategic direction, and holds ultimate accountability for AI-related decisions, so their engagement ensures the appetite aligns with corporate strategy and fiduciary duties. Legal and Compliance officers (D) are equally critical because they translate regulatory obligations (e.g., GDPR, EU AI Act, sector-specific rules), contractual duties, and ethical/legal constraints into concrete risk thresholds and controls, ensuring the stated appetite is lawful and enforceable. The building security personnel (A) focus on physical premises protection, not enterprise AI risk governance, so they are not key stakeholders here.

The office cleaning staff (C) have no role in defining strategic risk tolerance. The company's social media followers (E) are external parties whose opinions may inform reputation considerations but who hold no governance authority over risk appetite.

Exam trap

AAIR often tests the misconception that AI risk appetite is set by technical teams or external parties, when governance frameworks require Board-level ownership supported by Legal/Compliance, not facilities staff or social media audiences.

Full explanation →
23
Multi-Selecthard

Which THREE risks are associated with 'Automated Deployment' of AI models without a human-in-the-loop?

Select 3 answers
A.Rapid propagation of faulty or biased models to production.
B.The deployment takes too long to run.
C.Increased risk of security vulnerabilities being overlooked in the deployment config.
D.Lack of human intuition to catch edge-case failures.
E.Increased use of disk space.
AnswersA, C, D

Automated deployment pushes model artefacts straight to production without human review, so a faulty or biased model reaches all users almost instantly and at scale. The absence of a human-in-the-loop gate removes the checkpoint that would otherwise catch defects before release, enabling rapid propagation.

Why this answer

Option A is correct because automated deployment pipelines can push a faulty or biased model artifact straight to production without any manual review gate, so errors propagate at machine speed across all serving instances. Option C is correct because without a human reviewing the deployment configuration (IAM roles, exposed endpoints, secrets handling, container/image provenance), misconfigurations and security vulnerabilities can be merged and applied automatically. Option D is correct because human intuition and contextual judgment are often what catch edge-case or out-of-distribution failures that automated tests and metrics do not cover, and removing the human-in-the-loop eliminates that safeguard.

Option B is not a risk of automation — automated deployment is generally faster, not slower, than manual release. Option E is not a risk of automated deployment; disk space consumption depends on model size and artifact retention, not on whether a human approves the deployment.

Exam trap

AAIR often tests whether candidates can separate genuine governance/safety risks of automation from unrelated operational concerns like speed or disk space, so the trap is selecting distractors that sound like deployment problems but are not risks of removing human oversight.

Full explanation →
24
MCQmedium

Which role is primarily responsible for ensuring that AI-generated content adheres to intellectual property risk policies?

A.Chief Legal Officer
B.System Administrator
C.Product Manager
D.Data Engineer
AnswerA

The Chief Legal Officer owns intellectual property risk policy, defining how AI-generated content must respect copyright, licensing and IP ownership. This role sets the compliance boundaries that AI governance and content teams then operationalise through controls and review.

Why this answer

The Chief Legal Officer is primarily responsible for ensuring that AI-generated content adheres to intellectual property risk policies, as they oversee legal compliance and IP strategy. This role ensures that AI outputs do not infringe on copyrights, trademarks, or other IP rights. Other roles lack the legal expertise and authority.

Exam trap

AAIR often tests the assignment of IP risk responsibility, and candidates may incorrectly choose technical roles like Data Engineer or Product Manager, overlooking the legal expertise required.

How to eliminate wrong answers

Option B is wrong because a System Administrator manages IT infrastructure and has no responsibility for IP policy compliance. Option C is wrong because a Product Manager focuses on product development and may not have the legal expertise to enforce IP policies. Option D is wrong because a Data Engineer builds data pipelines and is not accountable for legal IP risk.

Full explanation →
25
MCQmedium

An enterprise is establishing an AI Governance Committee. Which organizational structure best ensures that risk management is integrated into the AI development lifecycle?

A.Centralized control under the Chief Data Officer
B.External advisory board oversight
C.Cross-functional membership including Legal, Security, and Engineering
D.Automated AI risk assessment tools alone
AnswerC

Including Legal, Security, and Engineering gives the committee authority over compliance, threat, and build decisions simultaneously, embedding risk review into each development stage. This satisfies the stem's integration requirement, which a single-function or purely executive body cannot achieve.

Why this answer

A cross-functional AI Governance Committee including Legal, Security, and Engineering ensures that legal/regulatory, security/privacy, and technical feasibility perspectives are all represented in AI risk decisions. This integration embeds risk management into the development lifecycle because engineering is at the table alongside the functions that define constraints and obligations, enabling risks to be identified and mitigated during design rather than after deployment.

Exam trap

AAIR often tests whether candidates equate governance with either a single executive owner or a purely external/automated mechanism — the trap is missing that effective AI governance requires cross-functional, lifecycle-embedded membership.

How to eliminate wrong answers

Option A is wrong because centralized control under the Chief Data Officer concentrates authority in one function, which can silo risk management and exclude legal, security, and engineering perspectives needed for lifecycle integration. Option B is wrong because an external advisory board provides independent oversight but is not embedded in the development lifecycle — it lacks the day-to-day integration needed to influence design decisions. Option D is wrong because automated AI risk assessment tools alone cannot replace human governance; tools detect issues but do not establish accountability, policy, or cross-functional decision-making.

Full explanation →
26
MCQhard

An organization discovers that their AI model exhibits bias toward a specific demographic. What is the most important step in the Incident Response process?

A.Suspend the model's production use until the issue is investigated and remediated.
B.Retrain the model immediately.
C.Delete all training data related to the demographic.
D.Issue a public press release.
AnswerA

Suspending production use immediately halts further biased decisions and affected users while root cause is investigated, containing harm before remediation. This satisfies the incident response priority of containment, which precedes analysis and correction in the lifecycle.

Why this answer

When an AI model is found to exhibit harmful bias, the immediate priority in incident response is containment — suspending production use prevents ongoing discriminatory decisions, legal exposure, and harm to affected individuals while the root cause is investigated. This mirrors the 'containment before eradication' principle in standard IR frameworks (NIST SP 800-61). Retraining or deleting data without first stopping the model leaves the biased system live and compounding harm.

Exam trap

AAIR often tests the instinct to 'fix it fast' — candidates pick retraining or data deletion because those feel proactive, but IR doctrine requires containment and evidence preservation before remediation.

How to eliminate wrong answers

Option B is wrong because retraining immediately skips containment and root-cause analysis; without understanding why the bias arose (data, labels, features, or objective function), retraining may reproduce or worsen the bias. Option C is wrong because deleting training data destroys forensic evidence needed for the investigation and may itself violate data retention or legal hold obligations. Option D is wrong because a public press release is a communications action, not an incident response step, and issuing one before containment and facts are established can amplify reputational and legal damage.

Full explanation →
27
MCQeasy

During the AI lifecycle, when should a 'Data Quality' assessment be performed to minimize long-term risk?

A.After model deployment in production.
B.During the final model sign-off.
C.During the decommission phase.
D.During the data ingestion and preprocessing stage.
AnswerD

Data quality assessment belongs at ingestion and preprocessing, where source data is profiled, cleaned and validated before training. Catching missing values, bias or schema drift here prevents contaminated datasets propagating through model development, satisfying the minimise-long-term-risk constraint.

Why this answer

Data quality assessment must happen during data ingestion and preprocessing, because that is when data is collected, cleaned, validated, and transformed before it reaches model training. Catching quality issues here prevents garbage-in-garbage-out propagation and is far cheaper than remediating after deployment. This early-stage control minimizes long-term risk across the entire AI lifecycle.

Exam trap

AAIR often tests the misconception that data quality is a late-stage or post-deployment concern, when the highest-leverage control point is ingestion and preprocessing.

How to eliminate wrong answers

Option A is wrong because assessing data quality after production deployment is too late — the model is already making decisions on flawed data and remediation is costly. Option B is wrong because final model sign-off is a late-stage checkpoint where data issues have already been baked into the trained model. Option C is wrong because the decommission phase is the end of the lifecycle and has no bearing on preventing data quality risk during development.

Full explanation →
28
Multi-Selecthard

Which THREE factors should be monitored to detect 'Model Drift' in the deployment phase?

Select 3 answers
A.The distribution of prediction outputs.
B.The temperature of the server room.
C.The current market price of GPU hardware.
D.The confidence score of the model predictions.
E.The distribution of incoming input features.
AnswersA, D, E

Monitoring the distribution of prediction outputs detects drift because shifts in input data or concept change the spread and frequency of predicted classes or values. Tracking this axis reveals degradation even when ground-truth labels are unavailable in the deployment phase.

Why this answer

Model drift detection in the deployment phase focuses on changes in data and model behavior over time. Option A, the distribution of prediction outputs, is correct because a shift in the distribution of predictions (e.g., a classifier suddenly predicting one class far more often) is a direct symptom of model drift, often caused by changing input data or degraded decision boundaries. Option D, the confidence score of the model predictions, is correct because declining or abnormally shifting confidence scores indicate the model is becoming less certain and less aligned with the data it was trained on, a common signal of drift.

Option E, the distribution of incoming input features, is correct because monitoring feature distributions (e.g., via statistical tests like PSI or KL divergence) detects data drift, which is the primary driver of model drift in production. Option B, the temperature of the server room, is incorrect because it is a physical infrastructure concern unrelated to statistical model behavior. Option C, the current market price of GPU hardware, is incorrect because it is a procurement/cost factor with no bearing on whether a deployed model's performance is degrading.

Full explanation →
29
MCQhard

A retail bank's AI risk committee is defining how third-party foundation model providers should be governed across 40 planned use cases. Legal notes that several providers reserve the right to change model versions with limited notice, and business units want to avoid re-approval delays for every update. Which approach BEST balances risk oversight with delivery speed?

A.Delegate all provider oversight to procurement and rely on contractual terms as the primary risk control.
B.Tier use cases by risk and data sensitivity, define pre-approved model versions and change classes per tier, and require impact assessment plus escalation only when a change crosses defined thresholds.
C.Standardize on a single foundation model provider to reduce the number of version-change events the committee must review.
D.Require full AI risk committee re-approval for every provider model version change before any production use.
AnswerB

Risk-based tiering lets the bank pre-authorize low-risk use cases and minor change classes, such as non-behavioral performance updates, while reserving committee escalation for changes that cross thresholds like new modalities, altered data retention, or use in credit decisions. It directly addresses the vendor's limited-notice practice by defining in advance which changes are acceptable and which trigger reassessment, preserving both oversight and delivery speed.

Why this answer

The bank needs governance that is proportionate and fast. Tiering use cases by risk and data sensitivity, pre-approving specific model versions and classes of change, and escalating only when a change crosses defined thresholds gives the committee control over material risk while letting low-impact updates proceed. This answers the vendor's limited-notice practice with pre-defined acceptance criteria, rather than serial re-approval or reliance on contracts alone.

Exam trap

The trap here is treating every model version update as equally risky and demanding identical re-approval, or conversely treating contracts as sufficient oversight.

Full explanation →
30
Multi-Selectmedium

Which TWO of the following are key components of a robust AI lifecycle risk management program?

Select 2 answers
A.Continuous model performance and data quality monitoring.
B.Using only open-source libraries.
C.Strict adherence to a single coding language.
D.Comprehensive versioning of data, code, and model artifacts.
E.Increasing the number of developers on the team.
AnswersA, D

Continuous monitoring of model performance and input data quality detects drift, degradation and data issues throughout deployment, enabling timely remediation. This ongoing vigilance is a core component of lifecycle risk management, satisfying the requirement for sustained oversight beyond initial validation.

Why this answer

Option A is correct because continuous monitoring of model performance and data quality is essential for detecting drift, degradation, and data issues throughout the AI lifecycle, enabling timely retraining or remediation. Option D is correct because comprehensive versioning of data, code, and model artifacts provides traceability, reproducibility, and auditability, which are foundational to managing risk across the AI lifecycle. Options B, C, and E are not key components of AI lifecycle risk management: using only open-source libraries, adhering to a single coding language, and increasing team size do not by themselves address risk controls, governance, or model reliability.

Exam trap

AAIR often tests whether candidates can distinguish genuine risk-management controls from superficial engineering preferences — the trap is selecting options that sound like 'best practices' (open source, one language, more staff) but do not actually mitigate AI lifecycle risk.

Full explanation →
31
Multi-Selectmedium

Which TWO of the following governance actions should a Chief Risk Officer (CRO) implement to align AI risk with the COSO Enterprise Risk Management (ERM) framework?

Select 2 answers
A.Establish a cross-functional AI Risk Committee
B.Assign AI risk ownership to individual data scientists
C.Disable all open-source AI libraries
D.Mandate daily model training cycles
E.Define AI-specific risk appetite statements within the ERM policy
AnswersA, E

COSO ERM requires governance and oversight structures spanning the entity. A cross-functional AI Risk Committee embeds AI risk into that oversight layer, giving the CRO a formal body to assign accountability, escalate issues and integrate AI into existing ERM reporting lines.

Why this answer

Option A is correct because the COSO ERM framework emphasizes governance and operating structures, and a cross-functional AI Risk Committee establishes the oversight body needed to integrate AI risk across the enterprise and align it with strategic objectives. Option E is correct because COSO ERM requires defining risk appetite and tolerance, so AI-specific risk appetite statements embedded in the ERM policy ensure AI risks are governed within the organization's overall risk posture. Option B is incorrect because assigning AI risk ownership solely to individual data scientists fragments accountability and ignores COSO's principle of enterprise-level ownership and oversight.

Option C is incorrect because disabling all open-source AI libraries is a blunt technical control, not a governance action, and would impede innovation without addressing risk alignment. Option D is incorrect because mandating daily model training cycles is an operational/technical decision unrelated to COSO ERM governance alignment.

Exam trap

AAIR often tests the distinction between governance-level actions (committees, risk appetite) and operational/technical actions (disabling libraries, daily training), tempting candidates to select hands-on controls over framework alignment.

Full explanation →
32
MCQeasy

During the 'Monitoring' phase, which activity is most critical for identifying model 'feedback loops'?

A.Tracking the distribution of prediction outputs over time for systemic shifts.
B.Validating the hardware security keys.
C.Checking for system uptime.
D.Analyzing the training pipeline latency.
AnswerA

Tracking the prediction output distribution over time detects systemic drift in what the model emits, which is the mechanism by which feedback loops manifest: outputs become inputs, skewing future predictions. This satisfies the Monitoring phase constraint of catching loops before they compound.

Why this answer

Feedback loops occur when a model's own predictions influence future input data, causing the model's output distribution to drift in a self-reinforcing way. Tracking the distribution of prediction outputs over time is the most direct way to detect these systemic shifts, because a feedback loop manifests as a gradual, self-amplifying change in what the model predicts. This is a monitoring-phase activity, distinct from training or infrastructure checks.

Exam trap

The trap is picking an infrastructure or pipeline metric (uptime, latency, security keys) because it sounds like 'monitoring' — the exam wants you to recognize that feedback loops are a model-behaviour phenomenon detectable only through output/input distribution tracking.

How to eliminate wrong answers

Option B is wrong because hardware security key validation is an infrastructure/security control unrelated to detecting model behaviour feedback loops. Option C is wrong because system uptime is an availability metric and says nothing about whether predictions are drifting or self-reinforcing. Option D is wrong because training pipeline latency measures the speed of the training process, not the runtime behaviour of the deployed model where feedback loops actually occur.

Full explanation →
33
Multi-Selecteasy

Which TWO of the following are considered 'High-Risk' AI use cases that require enhanced oversight by the Governance Committee?

Select 2 answers
A.Basic spell-checking software.
B.AI-powered biometric identification for facility access.
C.Automated recruitment and hiring filtering tools.
D.Email auto-complete features.
E.AI systems used for internal cafeteria menu suggestions.
AnswersB, C

Biometric identification for facility access processes special-category biometric data to make consequential decisions about individuals, placing it in the high-risk category under AI governance frameworks. It therefore requires enhanced oversight, impact assessment and Governance Committee review before deployment.

Why this answer

Option B is correct because AI-powered biometric identification for facility access processes sensitive biometric data and controls physical security, making it a high-risk use case that demands enhanced oversight by the Governance Committee. Option C is correct because automated recruitment and hiring filtering tools directly affect individuals' employment opportunities and can introduce bias or discrimination, which places them in the high-risk category requiring stricter governance. Options A and D are not high-risk since basic spell-checking and email auto-complete are low-impact productivity features with minimal risk to rights or safety.

Option E is also not high-risk because internal cafeteria menu suggestions have negligible impact on individuals and pose no significant compliance or ethical concerns.

Exam trap

AAIR often tests whether candidates can distinguish consequential, rights-affecting AI use cases from benign productivity tools, luring them into over- or under-classifying based on technical sophistication rather than impact.

Full explanation →
34
Multi-Selecthard

A fintech company is preparing its annual AI risk report for the board. The Chief Risk Officer wants the report to demonstrate that the AI risk program is operating effectively, not merely listing AI use cases. Which two elements most directly provide evidence of an effective operating AI risk program? (Choose two.)

Select 2 answers
A.The total number of employees who have completed general security awareness training.
B.A summary of the AI risk team's organizational chart and reporting lines.
C.A list of AI vendor logos and product names used across the organization.
D.Documented remediation actions and closure rates for issues raised by AI risk assessments and incident reviews.
E.A current inventory of AI systems with assigned risk tiers, accountable owners, and last assessment dates.
AnswersD, E

Tracking remediation actions and their closure rates shows that identified risks lead to concrete corrective work and that the program follows through. Closure rates over time give the board a measurable indicator of program responsiveness and discipline, distinguishing an operating program from one that only produces assessment documents.

Why this answer

Evidence of an operating program comes from artifacts that change over time in response to activity. A live AI inventory with tiers, owners, and assessment dates proves continuous identification and accountability, while remediation tracking with closure rates proves that findings translate into corrective action. Together they show the program detects, assigns, and resolves AI risk, which is what the board needs to judge effectiveness.

Exam trap

The trap here is mistaking structural or activity-count artifacts, such as org charts and generic training completion, for evidence that the AI risk program is actually operating and producing outcomes.

Full explanation →
35
Multi-Selectmedium

Which THREE items are necessary for a comprehensive 'AI Governance Policy'?

Select 3 answers
A.Standard Operating Procedures (SOPs) for model lifecycle management.
B.A list of every single line of code in the production models.
C.The personal home addresses of all developers.
D.AI risk classification framework (e.g., Low, Medium, High).
E.Definition of roles and responsibilities for AI risk oversight.
AnswersA, D, E

SOPs translate governance principles into repeatable, auditable steps covering model development, validation, deployment, monitoring and retirement. Without documented lifecycle procedures, an AI governance policy remains aspirational and cannot demonstrate consistent, traceable control across every model the organisation operates.

Why this answer

Option A is correct because a comprehensive AI Governance Policy must include Standard Operating Procedures (SOPs) that define the repeatable processes for model lifecycle management, covering stages such as data collection, training, validation, deployment, monitoring, and retirement, which ensures consistency and accountability. Option D is correct because an AI risk classification framework (e.g., Low, Medium, High) is essential for tiering AI systems by potential impact, so that controls, review rigor, and approval levels are proportionate to the risk each system presents. Option E is correct because clearly defining roles and responsibilities for AI risk oversight establishes ownership and accountability — for example, who approves models, who monitors drift, and who handles incidents — which is a foundational element of any governance policy.

Option B is not appropriate because listing every single line of code is an impractical and unnecessary level of detail for a governance policy; governance addresses principles, controls, and processes rather than source code inventories. Option C is not appropriate because collecting developers' personal home addresses is irrelevant to AI governance and would raise serious privacy and data protection concerns.

Exam trap

AAIR often tests whether candidates can distinguish governance artifacts (policies, frameworks, role definitions) from operational or irrelevant items (source code listings, personal data) — distractors like 'every line of code' sound thorough but are impractical.

Full explanation →
36
MCQeasy

What is 'Shadow AI'?

A.A marketing term for new AI tools
B.A feature in a programming language
C.AI used without formal approval or oversight
D.An AI model that only works in the dark
AnswerC

Shadow AI covers any AI tool, model or service deployed by staff outside sanctioned procurement and governance channels, so it escapes security review, data-handling controls and monitoring. The defining constraint is absence of formal approval or oversight, which is exactly what this option states.

Why this answer

'Shadow AI' refers to AI systems, tools, or models that are used within an organization without formal approval, oversight, or governance. This includes employees using public generative AI tools, deploying unvetted models, or integrating AI into workflows outside the sanctioned IT/security review process. It mirrors the concept of 'Shadow IT' but applied specifically to AI, and it represents a significant governance and compliance risk.

Exam trap

AAIR often tests... the confusion between 'Shadow AI' and legitimate emerging AI terminology, or candidates overthinking the term and selecting a technical-sounding but incorrect definition.

How to eliminate wrong answers

Option A is wrong because 'Shadow AI' is not a marketing term — it is a recognized governance and security risk category. Option B is wrong because it is not a programming language feature; it describes an organizational behavior, not a technical capability. Option D is wrong because it is a literal misinterpretation of the word 'shadow' — the term has nothing to do with lighting conditions or model operation in darkness.

Full explanation →
37
Multi-Selecthard

Which TWO aspects of 'Data Governance' are critical for AI risk management?

Select 2 answers
A.Data access control and security
B.The number of pens in the supply closet
C.The speed of the local internet
D.The interior design of the office
E.Data quality and consistency
AnswersA, E

Data access control and security govern who and what may read or modify training and inference data, directly constraining unauthorised exposure, poisoning and leakage. Enforcing identity-based permissions and encryption underpins every other AI risk control, making it a foundational governance requirement.

Why this answer

Data governance for AI risk management hinges on controlling who can access data and ensuring that data is trustworthy. Option A, data access control and security, is correct because AI systems depend on sensitive datasets, and governance must enforce authentication, authorization, encryption, and audit logging to prevent unauthorized access, data leakage, and tampering that could corrupt model behavior or violate privacy regulations. Option E, data quality and consistency, is correct because AI models learn directly from data; inaccurate, incomplete, duplicated, or inconsistent data leads to biased predictions, poor generalization, and unreliable outputs, so governance must define data standards, validation rules, and stewardship processes.

The remaining options are irrelevant to AI risk management: the number of pens in the supply closet (B), the speed of the local internet (C), and the interior design of the office (D) are office logistics or aesthetics and have no bearing on data governance or AI risk controls.

Exam trap

AAIR often tests whether candidates can distinguish genuine data governance dimensions (access, quality, lineage, retention) from irrelevant operational or facilities items — the trap is over-selecting plausible-sounding but non-governance options.

Full explanation →
38
Multi-Selecthard

Which THREE artifacts should be reviewed during a post-incident AI risk analysis?

Select 3 answers
A.The employee training certificates.
B.The building access logs for the data center.
C.The original Model Card and documentation.
D.The raw input and prediction output logs.
E.The model versioning and lineage logs.
AnswersC, D, E

The Model Card documents intended use, training data, performance metrics, and known limitations, so reviewing it reveals whether the incident stemmed from documented weaknesses or out-of-scope deployment. This grounds the post-incident AI risk analysis in the model's declared design and constraints.

Why this answer

The original Model Card and documentation (C) must be reviewed because it defines the model's intended purpose, training data characteristics, performance metrics, and known limitations, which are essential to determine whether the incident stemmed from out-of-scope use or undocumented behavior. The raw input and prediction output logs (D) are critical because they let analysts reconstruct the exact inference requests and responses surrounding the incident, revealing drift, adversarial inputs, or anomalous outputs that triggered the risk event. The model versioning and lineage logs (E) are also required to trace which model artifact, training dataset, and pipeline commit were actually serving at the time, since a regression or unauthorized model swap is a common root cause in AI incidents.

By contrast, employee training certificates (A) and building access logs for the data center (B) are general physical or HR compliance records that do not capture model behavior, data provenance, or inference-time evidence needed for an AI-specific post-incident analysis.

Exam trap

AAIR often tests whether candidates confuse physical/HR security artifacts with AI-specific lifecycle artifacts; the trap is selecting generic IT or facilities items (badges, certificates) instead of model-centric evidence like Model Cards, prediction logs, and lineage.

Full explanation →
39
MCQhard

In a cross-functional AI governance meeting, the Data Science team wants to deploy a new model, but the Legal team is concerned about data privacy. How should the AI Risk Manager resolve this?

A.Order the Data Science team to stop work immediately
B.Ask the CEO to decide
C.Conduct a joint risk assessment to identify potential mitigations
D.Ignore the Legal team's concerns
AnswerC

A joint risk assessment brings Data Science and Legal together to document the privacy concerns, evaluate likelihood and impact, and agree proportionate mitigations before deployment. This satisfies governance requirements for documented, cross-functional oversight rather than unilateral sign-off by either team.

Why this answer

The AI Risk Manager should facilitate a joint risk assessment that brings together Data Science and Legal to identify privacy risks and potential mitigations. This collaborative approach balances innovation with compliance, rather than unilaterally halting work or escalating to the CEO prematurely. It ensures both technical and legal perspectives inform the decision.

Exam trap

The trap is the false dichotomy between stopping work and ignoring concerns, which leads candidates to choose an extreme option instead of the collaborative risk assessment.

How to eliminate wrong answers

Option A is wrong because ordering an immediate stop to work is an extreme, unilateral action that ignores potential mitigations and may unnecessarily delay innovation. Option B is wrong because escalating to the CEO before attempting cross-functional resolution bypasses the AI Risk Manager's role and wastes executive time. Option D is wrong because ignoring Legal's concerns exposes the organization to regulatory and reputational harm and undermines the governance process.

Full explanation →
40
MCQhard

When defining KPIs for an AI Governance program, which metric is most predictive of long-term model robustness?

A.Number of users accessing the AI interface.
B.The frequency of model retraining triggered by performance drift alerts.
C.Total number of AI models currently in production.
D.The budget spent on GPU cloud computing resources.
AnswerB

Frequent drift-triggered retraining directly sustains robustness by refreshing the model against evolving data distributions, satisfying the long-term reliability constraint. Unlike static accuracy snapshots, this metric captures ongoing adaptation, revealing whether governance processes detect and correct degradation before it compounds.

Why this answer

The frequency of model retraining triggered by performance drift alerts is the most predictive metric of long-term robustness because it reflects the organization's ability to detect degradation and respond with corrective action. A governance program that tracks this metric demonstrates closed-loop monitoring and remediation, which sustains model performance over time as data distributions shift. This aligns with MLOps best practices for drift detection and continuous improvement.

Exam trap

AAIR often tests the confusion between vanity metrics (user counts, model counts, budget) and robustness indicators (drift-triggered retraining), so candidates must identify the metric that reflects closed-loop model health management.

How to eliminate wrong answers

Option A is wrong because the number of users accessing the AI interface measures adoption or popularity, not model robustness — a widely used model can still be fragile. Option C is wrong because the total number of models in production measures scale or portfolio size, not the health or robustness of any individual model. Option D is wrong because GPU cloud spend measures infrastructure cost, not model quality — high spend can indicate inefficiency rather than robustness.

Full explanation →
41
MCQhard

In the context of the Google Cloud Vertex AI Model Registry, what is the best approach to mitigate the risk associated with a model update that performs poorly on edge cases?

A.Implement traffic splitting with a shadow deployment.
B.Increase the training epoch count for the new version.
C.Directly replace the active production model version.
D.Update the training data to remove edge case records.
AnswerA

Shadow deployment mirrors live traffic to the updated model without serving its predictions, so poor edge-case behaviour is observed in production conditions while users still receive the stable model's responses. This satisfies the constraint of mitigating risk before any traffic shift.

Why this answer

Traffic splitting with a shadow deployment lets the new model version receive live traffic (or a copy of it) without affecting production responses, so its behaviour on edge cases can be observed before any real user impact. This directly mitigates the risk of a poorly performing update by gating promotion on measured performance. It is the standard safe-rollout pattern in Vertex AI Model Registry.

Exam trap

The trap is treating model updates as a code-deploy problem where 'just retrain harder' (more epochs) or 'just ship it' feels sufficient — the exam wants the safe-rollout pattern (shadow/canary) recognized as the risk control.

How to eliminate wrong answers

Option B is wrong because increasing epochs does not address edge-case failures and may worsen overfitting, making the model less robust on rare inputs. Option C is wrong because directly replacing the production model removes any safety net and exposes all users to the untested version's failures. Option D is wrong because removing edge-case records from training data hides the problem rather than solving it, and it degrades the model's coverage of legitimate rare scenarios.

Full explanation →
42
MCQmedium

In the context of the Three Lines of Defense, which function constitutes the 'Second Line'?

A.The Internal Audit department.
B.The Board of Directors.
C.The AI Development team.
D.The AI Risk Management and Compliance teams.
AnswerD

The Second Line provides independent oversight and challenge of risk decisions made by the business. AI Risk Management and Compliance teams perform exactly this oversight function, sitting between operational management in the First Line and internal audit in the Third Line.

Why this answer

In the Three Lines of Defense model, the Second Line consists of risk management and compliance functions that provide oversight, guidance, and challenge to the First Line (operational/business teams). For AI governance, the AI Risk Management and Compliance teams own frameworks, policies, and monitoring that ensure AI development and deployment align with risk appetite and regulatory requirements. They are independent from day-to-day operations but not fully independent like Internal Audit.

Exam trap

AAIR often tests whether candidates can correctly place functions in the Three Lines model, so they misassign Internal Audit (Third Line) or the Board (governing body) to the Second Line, or mistake a First Line development team for an oversight function.

How to eliminate wrong answers

Option A is wrong because Internal Audit constitutes the Third Line — it provides independent assurance to the board and must remain free from management influence, which is why it cannot sit in the Second Line. Option B is wrong because the Board of Directors sits above the Three Lines model as the governing body that receives assurance from all three lines; it is not itself a line of defense. Option C is wrong because the AI Development team is a First Line function — it owns and manages risk directly through the act of building and operating AI systems.

Full explanation →
43
Multi-Selecthard

When integrating AI risk into existing ERM, which THREE aspects of an AI model lifecycle must be audited to ensure compliance?

Select 3 answers
A.Personnel physical security badges
B.Office climate control settings
C.Model validation results
D.Incident response logs
E.Data lineage and provenance
AnswersC, D, E

Model validation results evidence that the model meets defined performance, fairness and robustness thresholds before deployment. Auditing these records confirms independent testing occurred, satisfying the lifecycle compliance requirement for AI risk integration within enterprise risk management.

Why this answer

Model validation results (C) must be audited because they document whether the AI model meets its stated performance, fairness, and robustness criteria before and after deployment, which is a core control for AI risk within ERM. Incident response logs (D) are essential because they provide evidence that AI-related failures, bias events, or security incidents were detected, escalated, and remediated according to the organization's risk response procedures. Data lineage and provenance (E) must be audited to verify that training, validation, and production data sources are authorized, traceable, and compliant with privacy and quality requirements, which directly affects model reliability and regulatory compliance.

Personnel physical security badges (A) and office climate control settings (B) are facility and environmental controls unrelated to the AI model lifecycle, so they do not belong in an audit of AI risk integration into ERM.

Exam trap

AAIR often tests whether candidates can distinguish AI lifecycle audit artifacts from generic IT/facilities controls; the trap is selecting physical-security or environmental items that sound like 'audit' topics but have no bearing on model risk.

Full explanation →
44
Multi-Selectmedium

A firm is establishing an AI Governance Committee. Which TWO groups should be represented to ensure comprehensive oversight?

Select 2 answers
A.Data Privacy Office
B.Public relations
C.Legal and Compliance
D.Hardware vendors
E.External marketing agencies
AnswersA, C

The Data Privacy Office contributes expertise on personal data handling, consent, retention and cross-border transfer, ensuring AI initiatives satisfy privacy regulation. Their inclusion satisfies comprehensive oversight by covering data-subject impact and privacy risk, which technical and business representatives alone would not adequately assess.

Why this answer

Legal/Compliance and Data Privacy are essential for addressing the regulatory and ethical risks inherent in AI deployments.

Full explanation →
45
Multi-Selecthard

When considering the 'Risk of Bias' in an AI system, which THREE components should be audited?

Select 3 answers
A.The physical location of the cloud data center.
B.The post-processing decisions made on the model outputs.
C.The composition and representativeness of the training dataset.
D.The number of coffee breaks the data science team takes.
E.The objective function or loss function used for model training.
AnswersB, C, E

Post-processing decisions directly shape how model outputs translate into outcomes, so auditing them exposes bias introduced by thresholds, filtering or business rules applied after inference. This satisfies the stem's requirement to audit bias risk across the system, since mitigation at the output stage is a distinct component from data and model training.

Why this answer

Option B is correct because bias can be introduced or amplified after the model produces its raw outputs, so auditing post-processing decisions (such as thresholding, ranking, filtering, or business rules applied to scores) is essential to detect and mitigate discriminatory outcomes. Option C is correct because the composition and representativeness of the training dataset directly determine whether the model learns patterns that generalize fairly; auditing for sampling bias, label bias, and underrepresentation of protected groups is a core part of a Risk of Bias assessment. Option E is correct because the objective or loss function encodes what the model optimizes for, and a poorly chosen loss (for example, optimizing overall accuracy on imbalanced data) can systematically disadvantage certain groups, so it must be audited.

Option A is not correct because the physical location of a cloud data center relates to data residency, sovereignty, or latency concerns, not to bias in model behavior. Option D is not correct because team break schedules have no bearing on algorithmic bias and are irrelevant to a Risk of Bias audit.

Exam trap

AAIR often tests whether candidates recognize that bias can be introduced at multiple stages, and a common mistake is to focus only on training data while ignoring the objective function and post-processing steps.

Full explanation →
46
MCQeasy

An organization is establishing an AI Risk Committee. Which stakeholder is most critical to include to ensure alignment between enterprise risk appetite and technical AI capabilities?

A.Chief Risk Officer (CRO).
B.Chief Information Security Officer (CISO).
C.Lead Cloud Architect.
D.Head of Human Resources.
AnswerA

The Chief Risk Officer owns the enterprise risk appetite and translates it into tolerances the committee applies. Including the CRO ensures AI technical controls are calibrated to that appetite, satisfying the alignment requirement between governance and engineering capability.

Why this answer

The Chief Risk Officer (CRO) is the executive who owns the enterprise risk appetite and the ERM framework, making them the most critical member to align AI risk governance with the organization's overall risk posture. Including the CRO ensures AI risks are scored, escalated, and mitigated using the same taxonomy and tolerance thresholds applied to financial, operational, and strategic risks. Without CRO involvement, AI risk decisions can drift from enterprise risk appetite.

Exam trap

AAIR often tests the distinction between enterprise risk ownership (CRO) and domain-specific security ownership (CISO) — candidates pick the CISO because AI security sounds like a security problem, but the question asks about enterprise risk appetite alignment.

How to eliminate wrong answers

Option B is wrong because the CISO, while essential for security-specific risks, owns a narrower domain (cyber and information security) and does not set enterprise-wide risk appetite — their scope does not cover model bias, regulatory, or business risk. Option C is wrong because a Lead Cloud Architect is a technical infrastructure role with no enterprise risk mandate and no authority over risk tolerance decisions. Option D is wrong because the Head of HR governs people and culture, not enterprise risk appetite, and has no role in AI risk scoring or mitigation strategy.

Full explanation →
47
MCQhard

You are managing a model that utilizes 'Online Learning'. What is the most critical risk requiring constant lifecycle vigilance?

A.Hardware failure in the training cluster.
B.Adversarial data poisoning and instability.
C.Data drift from batch updates.
D.Memory exhaustion due to high throughput.
AnswerB

Online learning updates weights continuously from live input, so attackers can inject poisoned samples that gradually corrupt the model, and rapid feedback loops can trigger runaway instability. Constant vigilance is required because no static validation checkpoint exists to catch either failure.

Why this answer

Online learning continuously updates model parameters from streaming data, which means the model's training distribution is directly influenced by whatever data arrives in production. This creates a critical, ongoing risk of adversarial data poisoning — attackers can inject malicious samples into the live stream to corrupt the model — as well as instability from feedback loops and non-stationary data. Constant lifecycle vigilance is required because the model can degrade or be subverted between reviews.

Exam trap

The trap is choosing generic infrastructure or performance risks (hardware, memory, batch drift) instead of recognizing that online learning's continuous data ingestion makes adversarial poisoning and instability the defining lifecycle risk.

How to eliminate wrong answers

Option A is wrong because hardware failure in the training cluster is an infrastructure resilience concern, not the distinctive risk of online learning, and is typically mitigated by redundancy rather than lifecycle vigilance. Option C is wrong because data drift from batch updates describes periodic retraining scenarios, whereas online learning updates continuously and its risk profile is dominated by poisoning and instability rather than batch-induced drift. Option D is wrong because memory exhaustion is a resource/scalability issue, not the core security and reliability risk unique to online learning.

Full explanation →
48
MCQmedium

When addressing 'Explainability' as an AI risk, which approach is most effective for a non-technical stakeholder?

A.Presenting the feature weights of a neural network.
B.Explaining the gradient descent optimization process.
C.Showing the mathematical formula used for loss reduction.
D.Providing counterfactual explanations like 'If your income had been $5k higher, the loan would have been approved'.
AnswerD

Counterfactual explanations state the minimal input change that would flip the outcome, expressing model behaviour in concrete, relatable terms without requiring mathematical or technical knowledge. This satisfies the need to make decisions intelligible to a non-technical stakeholder.

Why this answer

Counterfactual explanations are the most effective for non-technical stakeholders because they describe what would need to change in the input for a different outcome, using plain language and concrete scenarios. For example, 'If your income had been $5k higher, the loan would have been approved' is immediately understandable without any mathematical background. This approach aligns with GDPR Article 22 and the right to meaningful information about automated decisions.

Exam trap

AAIR often tests the difference between global and local explainability, and candidates frequently pick feature weights or SHAP values without considering the non-technical audience's need for actionable, plain-language reasoning.

How to eliminate wrong answers

Option A is wrong because feature weights of a neural network are abstract numerical values that non-technical stakeholders cannot interpret or act upon, and they do not convey decision rationale. Option B is wrong because gradient descent is an internal optimization algorithm, not an explanation of why a specific decision was made for a given individual. Option C is wrong because a loss reduction formula is a training-time mathematical construct that has no direct meaning for a stakeholder seeking to understand a single prediction.

Full explanation →
49
Multi-Selectmedium

Which THREE factors are essential to consider when determining the 'AI Risk Appetite' for an enterprise?

Select 3 answers
A.The current market share of the company.
B.The organization's tolerance for legal and reputational damage.
C.The complexity and autonomy level of the AI models being deployed.
D.The existing regulatory landscape and sector-specific compliance requirements.
E.Historical data volume.
AnswersB, C, D

Tolerance for legal and reputational damage directly bounds how much AI risk an enterprise will accept, satisfying the governance constraint that risk appetite must reflect consequences the board deems intolerable. Unlike technical metrics, this factor translates AI failures into enforceable liability and stakeholder trust thresholds, anchoring the appetite statement in measurable exposure limits.

Why this answer

Option B is correct because AI risk appetite must be calibrated to how much legal exposure and reputational harm the organization is willing to absorb, since these are the primary consequence categories of AI failures and directly bound the acceptable risk threshold. Option C is correct because the complexity and degree of autonomy of the deployed AI models determine the severity and likelihood of adverse outcomes, so higher autonomy or opacity requires a more conservative appetite. Option D is correct because the applicable regulatory landscape and sector-specific compliance obligations (for example, EU AI Act risk tiers, HIPAA, or financial-services rules) set non-negotiable limits that the risk appetite cannot exceed.

Option A is not essential because market share is a business-performance metric that does not define risk tolerance boundaries. Option E is not essential because historical data volume is an input to model development and data-governance decisions, not a determinant of how much AI risk the enterprise is willing to accept.

Exam trap

The trap is selecting business or technical metrics (market share, data volume) that sound relevant to AI but do not actually define risk tolerance — the exam tests whether you understand that risk appetite is about acceptable harm, not operational scale.

Full explanation →
50
MCQhard

When drafting a vendor management policy for AI, what is the most important clause to include?

A.Requirement for vendor to provide coffee
B.Requirement to use a specific font
C.Vendor must provide free hardware
D.Right-to-audit the vendor's AI training practices
AnswerD

A right-to-audit clause lets the organisation inspect the vendor's AI training data, model provenance and practices, providing enforceable oversight of risks such as bias, data misuse and regulatory non-compliance that other contractual terms cannot verify.

Why this answer

The most important clause in an AI vendor management policy is the right-to-audit the vendor's AI training practices, as it ensures transparency and accountability for the vendor's data handling, model development, and compliance with ethical standards. This clause allows the organization to verify that the vendor's AI systems meet regulatory and contractual requirements. Without it, the organization has limited visibility into potential risks.

Exam trap

AAIR often tests the importance of substantive clauses like right-to-audit versus trivial or irrelevant options, to see if candidates understand AI governance priorities.

How to eliminate wrong answers

Option A is wrong because requiring the vendor to provide coffee is trivial and irrelevant to AI risk management. Option B is wrong because mandating a specific font is a branding or formatting concern, not a substantive policy clause. Option C is wrong because free hardware is a commercial term, not a risk management control.

Full explanation →
51
MCQmedium

A practitioner is setting up a model monitoring service in AWS SageMaker Model Monitor. They observe that the ground truth data is significantly delayed. What action ensures risk identification remains effective?

A.Configure Data Quality Monitoring on input features only.
B.Disable monitoring until ground truth is available.
C.Manually force a model redeployment.
D.Reduce the sampling rate of the inference logs.
AnswerA

Delayed ground truth defeats real-time label comparison, so Data Quality Monitoring on input features detects distribution drift and anomalies in the incoming data itself, satisfying the need for effective risk identification without waiting for labels.

Why this answer

When ground truth labels are delayed, SageMaker Model Monitor cannot compute model quality metrics (accuracy, precision, etc.) because those require labels. The effective fallback is to enable Data Quality Monitoring on input features, which detects drift, missing values, and distribution shifts without needing labels, preserving risk identification in the interim.

Exam trap

AAIR often tests whether candidates know that Model Monitor's quality metrics depend on ground truth; the trap is choosing to disable or reduce monitoring when labels are late, instead of switching to label-free Data Quality monitoring.

How to eliminate wrong answers

Option B is wrong because disabling monitoring removes all visibility and guarantees blind spots during the exact period when drift is most likely. Option C is wrong because forcing a redeployment does not address the missing ground truth and may introduce new risk without evidence. Option D is wrong because reducing sampling lowers detection sensitivity and makes drift harder to catch, worsening risk identification rather than preserving it.

Full explanation →
52
MCQhard

When assessing the risk of AI-generated content, which factor is the most important for calculating 'Impact'?

A.The potential for reputational and legal consequences
B.The speed at which the content is generated
C.The color of the font used for the output
D.The number of GPUs used to create the content
AnswerA

Impact quantifies the severity of harm should the risk materialise. For AI-generated content, reputational damage and legal exposure — defamation, copyright infringement, regulatory breach — represent the most consequential outcomes, so they drive the Impact rating.

Why this answer

Impact in risk assessment is defined as the magnitude of consequences if a risk materializes; for AI-generated content, the most significant consequences are reputational damage and legal liability (defamation, copyright infringement, regulatory penalties, discrimination claims). These are the factors that determine severity in a risk matrix, whereas generation speed, font color, and GPU count are technical or cosmetic attributes with no bearing on harm magnitude.

Exam trap

AAIR often tests whether candidates correctly identify impact as consequence severity (reputational, legal) rather than confusing it with technical performance metrics like speed or resource usage.

How to eliminate wrong answers

Option B is wrong because generation speed is a performance metric, not an impact dimension — faster output does not increase the harm caused by inaccurate or harmful content. Option C is wrong because font color is a purely cosmetic attribute with no risk relevance whatsoever. Option D is wrong because the number of GPUs is a compute resource metric that affects cost and throughput, not the severity of consequences from the content itself.

Full explanation →
53
MCQeasy

Which of the following is an example of an 'AI Risk Metric'?

A.Budget spent on office supplies
B.Rate of false negatives in model output
C.Employee headcount in AI department
D.Number of coffee machines in the breakroom
AnswerB

A risk metric quantifies potential harm from AI system behaviour, and false-negative rate directly measures missed detections that could cause harm. It is a measurable, ongoing indicator of model reliability, satisfying the stem's requirement for an AI risk metric rather than a performance or business metric.

Why this answer

A risk metric quantifies the potential for harm or error in an AI system. The rate of false negatives directly measures how often the model misses positive cases, which is a core performance and risk indicator—especially in safety-critical applications like medical diagnosis or fraud detection. This metric helps assess the model's reliability and the risk of costly or dangerous omissions.

Exam trap

The trap here is confusing general business or operational metrics with AI-specific risk metrics; candidates might pick headcount or budget because they seem related to AI initiatives, but the question specifically asks for a metric that measures AI risk.

How to eliminate wrong answers

Option A is wrong because budget spent on office supplies is a financial accounting metric unrelated to AI model performance or risk. Option C is wrong because employee headcount in an AI department is a resource or staffing metric, not a measure of AI system risk. Option D is wrong because the number of coffee machines is a facility or workplace amenity metric, completely irrelevant to AI risk assessment.

Full explanation →
54
MCQmedium

What is the primary responsibility of a 'Model Owner' in an AI risk framework?

A.Writing all the Python code for the model
B.Approving company-wide bonuses
C.Fixing broken office furniture
D.Maintaining the risk profile and performance of the model
AnswerD

Model Owners own the model's risk profile and ongoing performance, satisfying the accountability requirement that someone continuously monitors accuracy, drift and risk exposure. This differs from governance bodies, which set policy, and from users, who consume outputs. The role therefore covers lifecycle monitoring and risk reporting rather than framework design or deployment engineering.

Why this answer

The Model Owner is accountable for the model's risk profile and performance throughout its lifecycle, including ensuring it operates within approved risk tolerances, monitoring for drift or degradation, and maintaining documentation for governance and audit. This role bridges technical teams and governance bodies, ensuring the model continues to meet business, ethical, and regulatory requirements. It is an accountability role, not necessarily a hands-on coding role.

Exam trap

AAIR often tests role confusion — candidates may equate 'Model Owner' with 'Model Developer' and pick the coding answer, missing that ownership is about risk accountability and lifecycle oversight, not hands-on implementation.

How to eliminate wrong answers

Option A is wrong because writing Python code is the responsibility of data scientists or ML engineers, not the Model Owner, whose role is oversight and accountability rather than implementation. Option B is wrong because approving company-wide bonuses is a finance or executive function entirely unrelated to AI risk management. Option C is wrong because fixing office furniture is a facilities task with no connection to model governance or risk frameworks.

Full explanation →
55
Multi-Selectmedium

When designing an AI Risk Reporting dashboard for senior management, which TWO of the following should be visualized?

Select 2 answers
A.Server uptime percentages.
B.Raw model prediction scores for every transaction.
C.Status of compliance certifications for high-risk models.
D.Aggregate risk score per business unit.
E.Employee training attendance logs.
AnswersC, D

Displaying compliance certification status for high-risk models shows senior management whether regulatory and internal obligations are currently met, satisfying the stem's reporting requirement. It highlights expired or pending attestations needing escalation, complementing aggregate risk scoring with governance assurance.

Why this answer

Option C is correct because senior management needs visibility into the governance posture of high-risk AI models, and tracking the status of compliance certifications (e.g., ISO/IEC 42001, SOC 2, or EU AI Act conformity assessments) directly supports regulatory oversight and accountability. Option D is correct because an aggregate risk score per business unit condenses model-level risk metrics into a decision-ready KPI that lets executives compare exposure, prioritize mitigation funding, and track risk appetite across the organization. Option A is not appropriate because server uptime is an infrastructure/availability metric that belongs in IT operations dashboards, not an AI risk reporting view for executives.

Option B is not appropriate because raw per-transaction prediction scores are far too granular and technical for senior management, and would create noise rather than actionable risk insight. Option E is not appropriate because employee training attendance is a tactical HR/compliance activity metric, not a strategic AI risk indicator for the board or executive committee.

Exam trap

AAIR often tests audience-appropriateness — candidates pick technically detailed options (raw scores, uptime) because they sound data-driven, but executive dashboards require aggregated, governance-relevant metrics, not operational noise.

Full explanation →
56
MCQeasy

What is the goal of an AI 'Model Inventory'?

A.To maintain visibility over all AI models and their associated risk profiles.
B.To automatically delete old models.
C.To provide marketing data to competitors.
D.To increase the speed of model deployment.
AnswerA

A Model Inventory provides a centralised register of every AI model in use, capturing ownership, purpose and risk classification. This directly satisfies the stem's goal by delivering the visibility and risk-profile oversight that governance, audit and regulatory reporting demand across the organisation's AI estate.

Why this answer

An AI Model Inventory is a centralized register that maintains visibility over all AI models in use, including their purpose, owner, data sources, version, and risk classification. Its primary goal is governance and risk management — you cannot govern what you cannot see. It enables oversight, audit, and lifecycle management rather than deployment speed or data sharing.

Exam trap

The trap is confusing the inventory's governance purpose (visibility and risk tracking) with operational side effects (deletion, speed, data sharing) — candidates may pick an action that the inventory enables rather than its actual goal.

How to eliminate wrong answers

Option B is wrong because automatically deleting old models is a lifecycle management action that may or may not be triggered by inventory data, but it is not the goal of the inventory itself — the inventory records and tracks models, it does not delete them. Option C is wrong because providing marketing data to competitors is nonsensical and contrary to the confidentiality purpose of an inventory; inventories are internal governance artifacts. Option D is wrong because increasing deployment speed is an operational benefit that might result from better visibility, but it is not the goal — the goal is risk visibility and accountability, and speed can even be constrained by governance.

Full explanation →
57
MCQhard

A multinational bank has assigned its AI governance responsibilities across the Three Lines of Defense. The second line has drafted an AI risk appetite statement, but the board wants assurance that the first line's AI use cases remain within that appetite. Which activity should the second line perform to provide this assurance while preserving its independence?

A.Directly approve or reject each first-line AI use case before deployment.
B.Independently review first-line AI risk assessments and challenge whether residual risk remains within the approved appetite.
C.Operate the AI monitoring dashboards and tune alert thresholds for model drift.
D.Commission an external audit of the AI governance program and report findings to the board.
AnswerB

Independent review and challenge of first-line AI risk assessments is a core second-line oversight activity. It provides assurance that residual risk is measured consistently against the board-approved appetite without taking ownership of the use case, preserving the objectivity required for effective Three Lines of Defense governance.

Why this answer

Second-line assurance over AI risk appetite is delivered through independent review and credible challenge of first-line assessments, not by taking over operational controls or audit functions. This keeps risk ownership with the first line while giving the board objective evidence that residual AI risk is being managed within the approved appetite.

Exam trap

The trap here is confusing oversight with ownership, assuming the second line must approve AI use cases rather than independently challenge the first line's risk assessments.

Full explanation →
58
MCQeasy

Which of the following is an example of an 'AI Risk' in a customer-facing service?

A.A chatbot providing inaccurate information to a customer.
B.The CEO deciding to change the corporate strategy.
C.The office printer running out of ink.
D.A scheduled maintenance window for the website.
AnswerA

Inaccurate chatbot responses are a risk because the generative model can produce plausible but false output, misleading customers and potentially causing harm or compliance breaches. This is an AI-specific risk, not a conventional infrastructure failure.

Why this answer

An AI risk in a customer-facing service is a harm or failure arising from the AI system's behavior — a chatbot providing inaccurate information is a classic example of hallucination/misinformation risk, which can mislead customers, cause financial or reputational harm, and create liability. It is directly attributable to the AI model's output quality and is a recognized risk category in AI governance frameworks.

Exam trap

AAIR often tests whether candidates can distinguish AI-specific risks (model behavior, data, output) from generic business or IT operational events — the distractors are mundane non-AI issues designed to look like risks.

How to eliminate wrong answers

Option B is wrong because a CEO changing corporate strategy is a business/strategic decision, not an AI system risk. Option C is wrong because a printer running out of ink is an operational facilities issue unrelated to AI. Option D is wrong because a scheduled maintenance window is planned IT operations, not an AI risk.

Full explanation →
59
Multi-Selectmedium

Which TWO factors are essential when documenting a model for risk management?

Select 2 answers
A.The model's intended use cases and operational boundaries.
B.The specific brand of coffee used by the developers.
C.The specific team member's personal hobbies.
D.The performance metrics and limitations observed during testing.
E.The names of all users who access the model.
AnswersA, D

Risk management depends on knowing where the model is deployed and what it is designed to do, since harms and mitigations differ by context. Documenting intended use cases and operational boundaries defines that scope, enabling assessors to judge whether deployed usage stays within validated limits.

Why this answer

Option A is correct because documenting a model for risk management must clearly state its intended use cases and operational boundaries, as these define the context in which the model is valid and where its risk exposure begins and ends. Option D is correct because recording the performance metrics and limitations observed during testing provides the quantitative evidence needed to assess residual risk, monitor degradation, and set appropriate controls or thresholds. Together, A and D establish both the scope of applicability and the empirical behavior of the model, which are foundational to any risk assessment.

Option B is incorrect because the developers' coffee brand is irrelevant to model risk. Option C is incorrect because personal hobbies of team members have no bearing on model risk documentation. Option E is incorrect because listing all users who access the model is an access-control or audit concern, not an essential factor for documenting the model itself for risk management.

Exam trap

AAIR often tests whether candidates confuse operational/technical documentation (use cases, metrics, limitations) with incidental metadata (user lists, developer trivia), so distractors are designed to look like 'documentation' without contributing to risk assessment.

Full explanation →
60
MCQhard

A company is scaling its AI initiatives across five business units. What is the most effective way to ensure consistent risk measurement?

A.Using different platforms for each business unit
B.Implementing a centralized AI risk taxonomy
C.Allowing each business unit to define its own risk metrics
D.Focusing only on the highest-risk model
AnswerB

A centralised AI risk taxonomy gives all five business units shared definitions, scoring scales and thresholds, so risk measurements remain comparable across initiatives. This satisfies the stem's consistency constraint when scaling governance beyond a single unit.

Why this answer

A centralized AI risk taxonomy provides a common language, scoring scale, and category structure that all five business units can apply consistently, enabling comparable risk measurement across the enterprise. Without a shared taxonomy, each unit's risk ratings become incommensurable, defeating the purpose of enterprise-wide AI risk oversight. This aligns with AAIR guidance that risk measurement must be standardized before it can be aggregated or governed.

Exam trap

AAIR often tests the misconception that decentralization or per-unit customization improves risk management, when in fact consistent measurement requires a shared, centralized taxonomy.

How to eliminate wrong answers

Option A is wrong because using different platforms per business unit fragments data and prevents aggregation or cross-unit comparison of risk metrics. Option C is wrong because letting each unit define its own metrics produces inconsistent scales and definitions, making enterprise-level risk reporting meaningless. Option D is wrong because focusing only on the highest-risk model ignores the portfolio view and leaves systemic or cumulative risks unmeasured across the other models.

Full explanation →
61
MCQhard

You are analyzing the risk of a generative AI implementation. Which metric most effectively measures 'model transparency' for a risk dashboard?

A.The number of unique contributors to the codebase.
B.The average latency of model inference.
C.Percentage of models deployed with a completed Model Card covering data provenance, limitations, and intended use.
D.The number of parameters in the neural network.
AnswerC

Model Cards document data provenance, limitations and intended use, giving the dashboard a measurable transparency metric. Tracking the percentage of deployed models with completed cards directly quantifies governance coverage, satisfying the requirement to measure model transparency as a risk indicator.

Why this answer

Model transparency is best measured by the percentage of deployed models that have a completed Model Card documenting data provenance, limitations, and intended use. Model Cards (Mitchell et al., 2019) are the industry-standard artifact for transparency, and a coverage metric directly quantifies how transparent the model portfolio is. This gives risk dashboards an actionable, auditable indicator tied to governance requirements.

Exam trap

AAIR often tests whether candidates can separate transparency metrics from performance, scale, or activity metrics, so the trap is picking a technical-sounding metric (parameters, latency) that measures capability rather than documented transparency.

How to eliminate wrong answers

Option A is wrong because the number of code contributors measures development activity or collaboration, not transparency of the model's behavior, data, or limitations. Option B is wrong because inference latency is a performance/operational metric (SLO), not a transparency indicator. Option D is wrong because parameter count is a measure of model size/capacity, not transparency — a model with billions of parameters can still be a complete black box.

Full explanation →
62
Multi-Selecthard

When implementing an AI Risk Management program, which THREE components are necessary for effective monitoring?

Select 3 answers
A.A requirement that all staff undergo coding training.
B.A mechanism for logging and auditing model outputs.
C.A daily newsletter for employees about AI advancements.
D.A process for human intervention or override.
E.Automated real-time monitoring of model performance and drift.
AnswersB, D, E

Logging and auditing model outputs creates the traceable record that monitoring depends on, capturing inputs, decisions and timestamps for later review. This satisfies the need to detect drift, investigate incidents and evidence compliance through continuous oversight.

Why this answer

Option B is correct because effective AI risk monitoring requires a mechanism for logging and auditing model outputs, which provides traceability, supports incident investigation, and enables detection of anomalies or policy violations over time. Option D is correct because a process for human intervention or override ensures that when a model behaves unexpectedly or produces harmful outputs, a human can step in to halt, correct, or escalate the issue, which is a core control in AI risk frameworks. Option E is correct because automated real-time monitoring of model performance and drift detects degradation, data drift, and concept drift as they occur, allowing timely remediation before risks materialize.

Option A does not belong because requiring all staff to undergo coding training is not a monitoring control and is irrelevant to observing model behavior. Option C does not belong because a daily newsletter about AI advancements is informational only and does not provide any monitoring, logging, or intervention capability.

Exam trap

AAIR often tests the difference between monitoring components and general awareness activities, tricking candidates into selecting training or communication options that feel like best practices but do not actually monitor AI risk.

Full explanation →
63
Multi-Selecthard

The AI Risk Governance Committee is defining the risk appetite for a new generative AI chatbot. Which THREE factors must be considered to align with organizational risk tolerance?

Select 3 answers
A.The speed of the development team's sprint cycles
B.Data privacy and residency requirements for training sets
C.The likelihood of model drift impacting output accuracy
D.The impact of potential hallucination on brand reputation
E.The total cost of GPU infrastructure
AnswersB, C, D

Data privacy and residency requirements constrain which training data may lawfully be used and where it resides. For a generative chatbot, these legal and contractual limits directly shape acceptable risk exposure, so the committee must weigh them against organisational risk tolerance.

Why this answer

Option B is correct because data privacy and residency requirements directly shape risk appetite: regulations like GDPR, HIPAA, or data-localization laws determine what training data may be used and where it can be stored, which bounds the acceptable risk for the chatbot. Option C is correct because model drift is a core AI risk factor — as data distributions shift, output accuracy degrades, so the committee must weigh the likelihood of drift against the organization's tolerance for degraded performance. Option D is correct because hallucination is a generative-AI-specific risk whose potential impact on brand reputation and customer trust must be assessed against the organization's risk tolerance.

Option A is not a risk-appetite factor; sprint cadence is a delivery-process metric, not a measure of AI risk exposure. Option E is also not a risk-tolerance factor; GPU infrastructure cost is a budgeting concern, not a determinant of acceptable risk.

Exam trap

AAIR often tests the confusion between operational cost/velocity metrics (GPU cost, sprint speed) and genuine risk appetite dimensions (privacy, drift, hallucination impact).

Full explanation →
64
MCQeasy

Which document should a practitioner review to determine the organization's threshold for acceptable AI model bias?

A.The vendor's Service Level Agreement (SLA).
B.The Enterprise Risk Appetite Statement.
C.The software's End User License Agreement (EULA).
D.The AI development team's sprint backlog.
AnswerB

The Enterprise Risk Appetite Statement formally records the organisation's willingness to accept risk, including the tolerance thresholds for AI model bias. It is the authoritative source for determining what level of bias is deemed acceptable.

Why this answer

The Enterprise Risk Appetite Statement is the governing document that defines how much risk the organization is willing to accept across its operations, including acceptable thresholds for AI model bias. It translates board-level risk tolerance into measurable limits that AI teams must design and validate against. Reviewing it tells a practitioner what bias level is tolerable versus what would breach organizational policy.

Exam trap

AAIR often tests whether candidates can distinguish governance documents (risk appetite, policy) from operational or legal documents (SLA, EULA, backlog), so the trap is selecting a document that sounds authoritative but does not define risk tolerance.

How to eliminate wrong answers

Option A is wrong because a vendor SLA defines service availability, support response times, and uptime commitments, not the organization's internal tolerance for model bias. Option C is wrong because an EULA is a legal license agreement governing software usage rights and liability, not a risk tolerance document. Option D is wrong because a sprint backlog is a tactical Agile artifact listing development tasks, not a governance document defining acceptable risk levels.

Full explanation →
65
Multi-Selecthard

Which THREE factors should be considered when assessing the 'Risk Level' of an AI application?

Select 3 answers
A.The sensitivity of the input data.
B.The criticality of the decision-making process being automated.
C.The complexity of the neural network architecture.
D.The potential impact on human rights and individual safety.
E.The color scheme of the user interface.
AnswersA, B, D

Input data sensitivity directly determines risk, because personal, special-category, or confidential data raises the consequences of breach, misuse, or model leakage. Higher sensitivity demands stricter controls and pushes the application toward a higher risk level under data protection frameworks.

Why this answer

Option A is correct because the sensitivity of the input data directly determines the potential harm if that data is exposed or misused, so higher-sensitivity data (e.g., PII, health, biometric data) raises the assessed Risk Level. Option B is correct because the criticality of the automated decision-making process reflects how consequential errors or failures would be, and automating high-stakes decisions (e.g., credit, medical triage) elevates risk. Option D is correct because potential impact on human rights and individual safety is a core risk dimension in AI risk frameworks, capturing harms such as discrimination, loss of liberty, or physical injury.

Option C does not belong because neural network architectural complexity is an implementation detail that does not by itself determine risk level. Option E does not belong because user interface color scheme is a cosmetic design choice with no bearing on AI risk assessment.

Exam trap

AAIR often tests whether candidates confuse technical complexity with risk level, leading them to select architecture-related options that sound sophisticated but are irrelevant to risk assessment.

Full explanation →
66
Multi-Selectmedium

Which THREE roles should have oversight authority in the AI Risk Program?

Select 3 answers
A.The company's social media influencer
B.Chief Technology Officer
C.Chief Risk Officer
D.The office receptionist
E.Head of Business Unit (where AI is used)
AnswersB, C, E

The CTO owns technology strategy and architecture, so they hold oversight authority over the technical controls, tooling and infrastructure underpinning the AI Risk Program. Their remit covers how AI systems are built and operated, satisfying the requirement for technology-level governance accountability.

Why this answer

The Chief Technology Officer (B) is correct because the CTO owns the technology strategy and AI systems architecture, so they must ensure AI risk controls are technically feasible, implemented, and aligned with the organization's engineering standards. The Chief Risk Officer (C) is correct because the CRO leads enterprise risk management, owns the risk framework and reporting to the board, and must ensure AI risks are identified, assessed, and mitigated consistently with other enterprise risks. The Head of Business Unit where AI is used (E) is correct because that leader is accountable for the business outcomes, operational use, and day-to-day risk decisions of the AI system, and must own residual risk within their unit.

The company's social media influencer (A) and the office receptionist (D) do not belong because neither role has the governance mandate, decision rights, or accountability for AI risk oversight in the organization.

Exam trap

AAIR often tests the roles that constitute oversight, and candidates may include irrelevant roles like influencers or receptionists, or omit business unit leaders, misunderstanding the need for business accountability.

Full explanation →
67
MCQmedium

Your organization is implementing an AI Risk Register. Which approach provides the most effective cross-functional coordination for identifying bias in a new HR recruitment AI?

A.Perform a technical audit on the training dataset distribution.
B.Automate all bias detection using internal library scripts without human review.
C.Update the AI policy to forbid the use of demographic attributes in training data.
D.Conduct a mandatory tabletop exercise involving HR, legal, and data science teams to simulate potential discriminatory outcomes.
AnswerD

A tabletop exercise forces HR, legal and data science to jointly simulate discriminatory outcomes, satisfying the stem's cross-functional coordination requirement. Unlike siloed reviews, it surfaces bias through shared scenario analysis, aligning risk identification across the disciplines that own recruitment decisions.

Why this answer

A tabletop exercise brings together HR, legal, and data science teams to simulate discriminatory outcomes, which is the most effective cross-functional coordination method because it surfaces bias risks that no single discipline can identify alone. HR understands job role requirements and historical hiring patterns, legal understands anti-discrimination law and liability, and data science understands model mechanics and data distributions. This collaborative simulation aligns with NIST AI RMF and ISO/IEC 42001 expectations for stakeholder engagement in AI risk identification.

Exam trap

AAIR often tests the misconception that bias identification is purely a technical/data-science task, when the exam expects recognition that cross-functional coordination (HR, legal, data science) is required for effective risk identification.

How to eliminate wrong answers

Option A is wrong because a technical audit on dataset distribution is a single-discipline (data science) activity that misses legal, ethical, and business context — it is necessary but not sufficient for cross-functional coordination. Option B is wrong because automating bias detection without human review removes the cross-functional judgment needed to interpret context-dependent bias and violates human-oversight principles. Option C is wrong because updating policy to forbid demographic attributes is a governance action, not a coordination mechanism — it also fails because proxy variables (e.g., zip code) can reintroduce bias even when protected attributes are removed.

Full explanation →
68
Multi-Selecthard

Which THREE components are necessary for effective cross-functional AI risk coordination?

Select 3 answers
A.Clearly defined RACI matrix for AI initiatives
B.Weekly team lunches
C.Standardized AI risk taxonomy
D.A centralized AI risk register
E.Unrestricted access to all production servers for all staff
AnswersA, C, D

A clearly defined RACI matrix assigns who is responsible, accountable, consulted and informed for each AI initiative, removing ambiguity over ownership. This satisfies the coordination requirement by ensuring every risk decision has a named owner across functions.

Why this answer

A clearly defined RACI matrix for AI initiatives (A) is necessary because it assigns explicit Responsible, Accountable, Consulted, and Informed roles across functions, eliminating ambiguity about who owns each AI risk decision and escalation path. A standardized AI risk taxonomy (C) is required so that legal, security, data science, and business teams classify and compare risks using the same categories, severity scales, and terminology, enabling consistent cross-functional assessment. A centralized AI risk register (D) is essential as the single source of truth that aggregates identified risks, owners, mitigations, and status, giving all functions shared visibility and traceability.

Weekly team lunches (B) are a social nicety that may improve rapport but are not a structural mechanism for risk coordination, and unrestricted access to all production servers for all staff (E) violates least-privilege and actually increases risk rather than enabling effective coordination.

Exam trap

AAIR often tests whether candidates can distinguish governance structure (RACI, taxonomy, register) from cultural or operational gestures (lunches, open access) — the trap is selecting options that sound collaborative or 'agile' but provide no accountability or traceability.

Full explanation →
69
MCQhard

A company is utilizing a third-party LLM service. Which risk is specifically heightened by this arrangement?

A.Data leakage and privacy violations
B.Server downtime
C.Model training cost inflation
D.Lack of software updates
AnswerA

With a third-party LLM service, prompts and embedded data leave the organisation's trust boundary and are processed on external infrastructure, so confidential information may be retained, logged or reused. This external processing arrangement is precisely what heightens data leakage and privacy violations compared with a self-hosted model.

Why this answer

Using a third-party LLM service means prompts, context, and potentially sensitive data are transmitted to and processed by an external provider, heightening data leakage and privacy violation risk. The organization loses direct control over data handling, retention, and secondary use, and may face regulatory exposure under GDPR, HIPAA, or similar regimes. This is the risk specifically amplified by the third-party arrangement.

Exam trap

AAIR often tests the misconception that operational concerns like downtime or updates are the main third-party LLM risk, when the specifically heightened risk is data leakage and privacy exposure.

How to eliminate wrong answers

Option B is wrong because server downtime is a general availability risk that applies to any hosted service and is not specifically heightened by using a third-party LLM versus self-hosting. Option C is wrong because training cost inflation is largely the provider's concern under a service model; the customer pays inference/API costs, not training costs, so this is not the heightened risk. Option D is wrong because lack of software updates is the opposite concern—third-party providers typically push updates continuously, and the real risk is uncontrolled model changes, not absence of updates.

Full explanation →
70
MCQeasy

Why is 'Data Lineage' important in an AI risk management program?

A.It helps ensure data quality and integrity
B.It makes user interfaces more attractive
C.It improves server load times
D.It reduces the amount of storage needed
AnswerA

Data lineage records the origin, movement and transformation of each dataset through the pipeline. This traceability underpins data quality and integrity by exposing errors, unauthorised changes and gaps, letting risk teams verify that AI training and inference data remains accurate and trustworthy.

Why this answer

Data lineage documents the origin, movement, transformation, and usage of data throughout its lifecycle, which is foundational to AI risk management because it enables traceability of data quality issues, bias sources, and compliance violations back to their root. Without lineage, an organization cannot verify that training data was lawfully obtained, representative, or free of contamination — making it impossible to assess or remediate data-related AI risks.

Exam trap

AAIR often tests whether candidates confuse data lineage (provenance and traceability) with unrelated concerns like UI design, performance, or storage optimization.

How to eliminate wrong answers

Option B is wrong because data lineage has no relationship to user interface aesthetics — UI design is a front-end concern unrelated to data provenance tracking. Option C is wrong because server load times are determined by infrastructure, caching, and query optimization, not by lineage metadata; lineage tracking may actually add minor overhead rather than improve performance. Option D is wrong because lineage metadata increases storage requirements (it records provenance information), not reduces them — the benefit is governance and traceability, not storage efficiency.

Full explanation →
71
MCQmedium

A manufacturing firm is adopting a 'Privacy by Design' approach for AI. How does this integrate with the AI Risk Governance framework?

A.Privacy becomes the responsibility of the vendor, not the company.
B.Privacy controls are integrated into the MLOps pipeline as automated gates.
C.Privacy is only reviewed after the model is deployed.
D.Privacy impact assessments are excluded to speed up model deployment.
AnswerB

Embedding privacy controls as automated gates in the MLOps pipeline enforces Privacy by Design at each stage, from data ingestion to model deployment, so privacy risk is governed continuously rather than reviewed once, integrating directly with the AI Risk Governance framework's controls.

Why this answer

Privacy by Design (PbD) is a proactive principle that requires embedding privacy protections into systems from the outset, not as an afterthought. In an AI Risk Governance framework, this translates to operationalizing privacy controls—such as data minimization, consent checks, and anonymization—as automated gates within the MLOps pipeline. This ensures that every model training, evaluation, and deployment step adheres to privacy requirements before progressing, making privacy an integral, continuous part of the AI lifecycle rather than a one-time or post-hoc review.

Exam trap

AAIR often tests the misconception that Privacy by Design is a post-deployment or vendor-only responsibility, when it actually requires proactive, organization-wide integration into every stage of the AI lifecycle.

How to eliminate wrong answers

Option A is wrong because Privacy by Design places responsibility on the organization deploying the AI, not solely on the vendor; the company remains accountable for compliance and ethical use. Option C is wrong because Privacy by Design mandates privacy considerations throughout the entire lifecycle, not just after deployment; post-deployment review is reactive and violates the proactive principle. Option D is wrong because Privacy Impact Assessments (PIAs) are a core component of Privacy by Design and AI risk governance; excluding them increases risk and non-compliance, directly contradicting the framework's goals.

Full explanation →
72
MCQmedium

A team notices that an AI model's performance decreases when the input data contains abbreviations that were not present in the training set. This is a risk associated with which lifecycle stage?

A.Infrastructure provisioning.
B.Data preprocessing and training design.
C.Monitoring phase.
D.Deployment phase.
AnswerB

Abbreviations absent from training data are a vocabulary and feature-engineering gap, so the risk originates in data preprocessing and training design, where tokenisation and normalisation decisions determine how unseen input forms are handled at inference.

Why this answer

Abbreviations absent from the training set are a vocabulary/coverage gap introduced during data collection and feature engineering. Because the model never learned representations for those tokens, it cannot generalize to them at inference — this is a data preprocessing and training design risk (B), not an infrastructure, monitoring, or deployment issue. The root cause is upstream of deployment: the training corpus did not represent the input distribution the model now sees.

Exam trap

AAIR often tests whether candidates attribute model-quality failures to the stage where they are observed (monitoring/deployment) rather than where they originate (data preprocessing and training design); the trap is picking 'monitoring' because that is when the drop is noticed.

How to eliminate wrong answers

Option A is wrong because infrastructure provisioning concerns compute, storage, and networking resources, not the linguistic coverage of the training data. Option C is wrong because monitoring detects the performance drop but does not cause it; the risk originates in training data design. Option D is wrong because deployment is about serving the model, not about whether the training vocabulary covered the input space.

Full explanation →
73
MCQmedium

A company is implementing 'Human-in-the-Loop' (HITL) for its AI decision system. What is the primary risk of relying on HITL?

A.The 'automation bias' or 'rubber stamping' of decisions by human reviewers.
B.It requires expensive API calls.
C.It causes the model to become slower.
D.It forces the model to use a specific programming language.
AnswerA

Reviewers facing high volumes of AI recommendations tend to accept them with minimal scrutiny, producing automation bias or rubber stamping. This erodes the independent verification HITL is meant to provide, so the human check becomes a formality rather than genuine oversight.

Why this answer

Human fatigue and cognitive bias can lead to poor oversight, potentially creating a false sense of security.

Full explanation →
74
MCQmedium

You are defining KPIs for an AI project in a regulated industry. Which metric best captures model reliability over time?

A.Model drift (data drift/concept drift)
B.Frequency of model retraining
C.Number of training iterations
D.Total cost of compute resources
AnswerA

Model drift measures divergence between training and live data distributions (data drift) or between learned relationships and outcomes (concept drift). Tracking it over time detects silent degradation in predictive reliability, which accuracy snapshots alone miss — essential for sustained oversight in regulated settings.

Why this answer

Model drift — encompassing data drift and concept drift — directly measures whether a model's predictive behavior remains aligned with the real-world distribution it was trained on. It is the canonical reliability KPI because it captures degradation over time before it manifests as business harm. Retraining frequency, iteration counts, and compute cost are operational metrics, not reliability indicators.

Exam trap

AAIR often tests the difference between reliability metrics (drift, performance degradation) and operational metrics (retraining frequency, iterations, cost) — candidates pick retraining frequency because it sounds like a proxy for reliability, but it measures activity, not outcome.

How to eliminate wrong answers

Option B is wrong because retraining frequency is an operational cadence metric — a model can be retrained often and still drift, or rarely and remain stable. Option C is wrong because the number of training iterations is a hyperparameter/tuning detail with no direct relationship to production reliability. Option D is wrong because total compute cost is a financial efficiency metric, not a measure of whether the model still performs correctly.

Full explanation →
75
MCQhard

In the context of the AI RMF, what is the significance of the 'Measure' function for risk management?

A.To identify which employees have access to the AI system.
B.To set the budget for future AI projects.
C.To provide a score for the developers' performance.
D.To evaluate the model's performance and risk outcomes against established benchmarks.
AnswerD

The Measure function quantifies AI risk by testing, analysing and tracking model performance and outcomes against defined benchmarks and metrics. This produces the evidence needed to assess whether risks fall within tolerance, satisfying the stem's requirement to evaluate performance and risk outcomes.

Why this answer

The Measure function of the AI RMF is responsible for evaluating AI systems' performance, trustworthiness characteristics, and risk outcomes against defined benchmarks and metrics. It provides the quantitative and qualitative evidence needed to determine whether risks are within tolerance and whether the system meets its intended purpose. This function operationalizes the risk assessment by turning identified risks into measurable indicators.

Exam trap

AAIR often tests the confusion between Measure (evaluating performance and risk outcomes) and Manage (implementing risk treatments), causing candidates to select an operational or governance activity that belongs to a different function.

How to eliminate wrong answers

Option A is wrong because identifying employee access is an access control and identity management activity, which falls under the Govern or Manage functions, not Measure. Option B is wrong because setting budgets for future AI projects is a financial planning activity unrelated to the Measure function's purpose of evaluating model performance and risk outcomes. Option C is wrong because scoring developer performance is a human resources function; the Measure function evaluates the AI system and its risks, not individual employee productivity.

Full explanation →

Page 1 of 3

Page 2

All pages

123

Practice AAIR by domain

Target a specific domain to shore up weak areas.

AI Risk Program ManagementAI Risk Governance And Framework IntegrationAI Lifecycle Risk Management
See all domains with question counts →