ISACA AAIR Advanced in AI Risk

Which NIST AI Risk Management Framework (AI RMF 1.0) function establishes organizational culture, accountability structures, and AI risk management policies?

• AMap
• BMeasure
• CGovern
• DManage

Under the EU AI Act, which article enumerates AI practices that are prohibited as posing unacceptable risk?

• AArticle 5
• BArticle 6
• CArticle 9
• DArticle 52

ISO/IEC 42001:2023 establishes requirements for which organizational capability?

• AAn AI management system (AIMS)
• BAI software testing
• CCloud security controls
• DData classification

Which standard provides guidance specifically for AI risk management aligned to ISO 31000?

• AISO/IEC 23894
• BISO/IEC 23053
• CISO/IEC 24029
• DISO/IEC 5338

The EU AI Act entered into force on which date, marking the start of its phased applicability schedule?

• AMay 21, 2024
• BAugust 1, 2024
• CFebruary 2, 2025
• DAugust 2, 2026

Under the EU AI Act, a general-purpose AI (GPAI) model is presumed to have systemic risk when training compute exceeds which threshold?

• A10^22 FLOPs
• B10^23 FLOPs
• C10^25 FLOPs
• D10^27 FLOPs

Which characteristic of trustworthy AI in NIST AI RMF 1.0 is most directly concerned with the AI system's resilience to adversarial inputs and unexpected conditions?

• AValid and Reliable
• BSafe
• CSecure and Resilient
• DExplainable and Interpretable

What is the maximum administrative fine under the EU AI Act for non-compliance with the prohibited AI practices in Article 5?

• AUp to €10 million or 2% of worldwide annual turnover
• BUp to €20 million or 4% of worldwide annual turnover
• CUp to €35 million or 7% of worldwide annual turnover
• DUp to €50 million or 10% of worldwide annual turnover

Which OECD AI Principles value-based principle most directly addresses the goal that AI systems should respect the rule of law, human rights, and democratic values?

• AInclusive growth, sustainable development and well-being
• BHuman-centered values and fairness
• CTransparency and explainability
• DRobustness, security and safety

GDPR Article 22 grants data subjects the right not to be subject to a decision based solely on which type of processing?

• AManual review by a data controller
• BAutomated processing, including profiling, that produces legal or similarly significant e…
• CAny algorithmic processing of personal data
• DCross-border data transfers

US Executive Order 14110 on Safe, Secure, and Trustworthy AI — issued by President Biden in October 2023 — was rescinded by which subsequent action?

• AOMB Memorandum M-24-10
• BExecutive Order 14179 issued in January 2025
• CThe CHIPS and Science Act
• DFTC Section 5 enforcement

OMB Memorandum M-25-21 (April 2025) primarily directs federal agencies to do which of the following?

• AProcure AI exclusively from the General Services Administration
• BAccelerate federal AI use, designate Chief AI Officers, and manage risks for high-impact…
• CBan use of generative AI by federal civilian agencies
• DReplace NIST AI RMF with a custom federal framework

Which of the following is a key purpose of an AI ethics committee within an enterprise AI risk program?

• ATo approve all AI vendor invoices
• BTo provide cross-functional review of AI use cases for ethical, legal, and societal risk
• CTo replace the Chief Information Security Officer
• DTo draft AI source code

Which document, published by the White House Office of Science and Technology Policy in October 2022, articulates non-binding rights-oriented principles for automated systems including safe and effective systems, algorithmic discrimination protections, data privacy, notice and explanation, and human alternatives?

• AThe Blueprint for an AI Bill of Rights
• BExecutive Order 13960
• CThe National AI Initiative Act
• DOMB Circular A-130

Under the EU AI Act, which Annex enumerates the high-risk use cases that fall under Article 6(2) (e.g., biometric identification, critical infrastructure, education, employment, essential services)?

• AAnnex I
• BAnnex II
• CAnnex III
• DAnnex IV

ISO/IEC 42005 provides guidance on which AI governance activity?

• AAI system impact assessment
• BAI training data labeling
• CAI cybersecurity controls
• DAI procurement contracts

In the financial services sector, which supervisory guidance from the US Federal Reserve and OCC most directly governs model risk management practices that apply to AI/ML models?

• AFFIEC IT Examination Handbook
• BSR 11-7 / OCC 2011-12 (Supervisory Guidance on Model Risk Management)
• CSOX Section 404
• DBasel III

Which FDA guidance describes the agency's evolving framework for regulating Artificial Intelligence and Machine Learning (AI/ML) Software as a Medical Device (SaMD), including the concept of a Predetermined Change Control Plan?

• A21 CFR Part 11
• BFDA AI/ML SaMD Action Plan and PCCP guidance
• CHIPAA Security Rule
• D21st Century Cures Act

An AI risk taxonomy used by an enterprise typically distinguishes between data risk, model risk, deployment risk, third-party risk, and societal risk. Which of the following is most accurately classified as societal risk?

• AA training dataset contains duplicated rows
• BA loan-approval model perpetuates historical lending discrimination at scale
• CA model overfits and fails on the validation set
• DA vendor's API exceeds latency SLOs

Which NIST publication accompanies AI RMF 1.0 to provide actionable guidance for governing and managing risks from generative AI?

• ANIST SP 800-53 Rev. 5
• BNIST AI 600-1 (Generative AI Profile)
• CNIST SP 800-37
• DNIST IR 8286

Which of the following best describes the 'limited-risk' category of AI systems under the EU AI Act?

• ASystems banned outright due to unacceptable risk
• BSystems subject to transparency obligations such as informing users they are interacting…
• CSystems requiring conformity assessment by a notified body
• DSystems exempt from any AI Act obligation

Which ISO/IEC standard specifies AI system life cycle processes from inception through retirement?

• AISO/IEC 5338
• BISO/IEC 5339
• CISO/IEC 23053
• DISO/IEC 22989

An AI policy framework should be integrated with which of the following enterprise policies to provide effective coverage?

• AAcceptable use, data governance, and cybersecurity policies only
• BAcceptable use, data governance, cybersecurity, privacy, and third-party risk policies
• COnly data classification and retention policies
• DOnly the records management policy

Within NIST AI RMF 1.0, the Map function primarily helps an organization to do which of the following?

• AEstablish AI policies and roles
• BEstablish the context, identify AI risks, and characterize their potential impact
• CQuantitatively benchmark AI performance
• DDecide which models to retire

Which of the following is a key cross-cutting characteristic that NIST AI RMF 1.0 considers essential for trustworthiness, particularly for generative AI outputs?

• AHardcoded thresholds
• BAccountable and Transparent
• CClosed-source training
• DSingle-vendor sourcing

Which of the following best describes a 'foundation model' in the context of AI risk governance?

• AAny small linear regression model used in finance
• BA large model trained on broad data at scale that can be adapted to many downstream tasks
• CA model that runs only on-device on edge hardware
• DA model used solely for tabular data

An AI risk register should at minimum capture which of the following attributes for each AI risk entry?

• ARisk description, owner, inherent likelihood/impact, controls, residual risk, and review…
• BOnly the model name and the data scientist's email
• COnly the model accuracy metric
• DOnly the regulator's name

Under the EU AI Act, who is the 'provider' of an AI system?

• AThe natural or legal person that develops or has an AI system developed and places it on…
• BOnly the cloud provider hosting the model
• COnly the end user
• DOnly the data subject

Which of the following is a recognized fairness metric that measures whether the true positive rate is equal across protected groups?

• ADemographic parity
• BEqual opportunity
• CEqualized odds
• DPredictive parity

Which of the following best describes the goal of a fundamental rights impact assessment (FRIA) under EU AI Act Article 27?

• AAudit a company's tax obligations
• BIdentify and mitigate risks of high-risk AI to people's fundamental rights when used by c…
• CEvaluate cost savings from AI deployment
• DCalculate carbon emissions

Which of the following is the most appropriate KRI (Key Risk Indicator) for monitoring a deployed credit-scoring model?

• ANumber of marketing emails sent
• BPopulation Stability Index (PSI) of input features over time
• CAverage employee tenure
• DCafeteria utilization

Which of the following is a primary goal of model documentation artifacts such as Model Cards (Mitchell et al., 2019)?

• ADocument model performance, intended use, limitations, ethical considerations, and evalua…
• BProvide source code for adversaries
• CReplace privacy notices
• DReplace the engineering on-call schedule

ISO/IEC 24029 series provides guidance on which AI quality attribute?

• ARobustness assessment of neural networks
• BAI cost accounting
• CAI marketing claims
• DAI hardware tolerances

An organization wants to formally certify its AI management system. Which standard provides certifiable requirements suitable for third-party audit?

• AISO/IEC 22989
• BISO/IEC 23053
• CISO/IEC 42001
• DNIST AI RMF 1.0

Which characteristic distinguishes 'systemic risk' GPAI models from baseline GPAI under the EU AI Act?

• AUse of GPU hardware
• BOpen-source license
• CHigh-impact capabilities, including the 10^25 FLOP training compute presumption
• DUse of supervised learning

Within an AI lifecycle (problem framing → data → model development → evaluation → deployment → monitoring → retirement), which stage most directly addresses representativeness, lawful basis, and labeling quality?

• AProblem framing
• BData
• CModel development
• DRetirement

Which type of attack manipulates the training data to corrupt the model so it behaves maliciously on specific inputs at inference time?

• AModel evasion
• BMembership inference
• CData poisoning (including backdoor attacks)
• DModel extraction

Concept drift in a deployed model refers to which phenomenon?

• ADecline in compute capacity
• BChange in the joint distribution P(X,Y) such that the relationship between inputs and the…
• CLoss of GPU drivers
• DEnd of model warranty

Which classification metric is most appropriate to monitor in a fraud-detection system where the positive class is highly imbalanced?

• AOverall accuracy
• BF1 score and precision-recall AUC
• CMean absolute error
• DR-squared

Which of the following best describes a 'membership inference' attack against a machine-learning model?

• AAn adversary submits crafted inputs to bypass moderation
• BAn adversary determines whether a specific record was part of the training data
• CAn adversary tampers with model weights at runtime
• DAn adversary purchases stolen GPUs

OWASP Top 10 for LLM Applications LLM01 addresses which class of risk?

• APrompt injection (direct and indirect)
• BInsecure output handling
• CTraining data poisoning
• DExcessive agency

An LLM-based assistant connects to a database via a tool plugin and is granted broad write access. Which OWASP Top 10 for LLM risk does this design choice most directly trigger?

• ALLM05 Supply Chain
• BLLM07 Insecure Plugin Design
• CLLM08 Excessive Agency
• DLLM10 Model Theft

MITRE ATLAS is best described as which of the following?

• AA penetration testing tool
• BA globally accessible knowledge base of adversary tactics and techniques against AI-enabl…
• CA financial-services trading framework
• DA consumer credit scoring model

Which technique uses a divergence-based statistic comparing a baseline distribution to current production data to detect data drift?

• AROC AUC
• BPopulation Stability Index (PSI) and KL divergence
• CCosine similarity of embeddings
• DPearson correlation of timestamps

Which mitigation is generally most effective against indirect prompt injection where an LLM agent processes untrusted web content?

• ATrust the model to refuse malicious instructions
• BTreat all retrieved content as untrusted, isolate tool execution, require human approval…
• CDisable logging to avoid exposing the prompt
• DMove the model to a different region