AAISM Practice Test 23 — 15 Questions

Question 1

A team is deploying a vision model for automated quality inspection on a factory floor. During red-teaming, they find that placing a small printed patch on a widget causes the model to classify defective items as passing. The team cannot retrain immediately. Which control most directly reduces the operational impact of this finding?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Introduce an independent inspection step, such as a secondary sensor or human sampling, that can override the model's pass decision.. A physical adversarial patch defeats the model's decision, and retraining is off the table in the near term. The pragmatic control is to ensure no single model output can release a defective item, which an independent sensor or human sampling accomplishes. Adversarial retraining, threshold tuning, and camera resolution all either take too long or fail to address the perturbation's effect on the classifier.

Answer

Increase the resolution of the inspection camera to capture finer surface detail.

Answer

Lower the model's classification threshold so that more items are flagged as defective.

Answer

Add the patch pattern to the training set as a labeled adversarial example and schedule retraining next quarter.

Question 2

A security analyst is reviewing controls for an AI system that uses a large language model (LLM) to generate responses based on user prompts. The analyst wants to mitigate the risk of prompt injection attacks that could cause the model to reveal sensitive information or perform unintended actions. Which TWO of the following controls are most effective for this purpose? (Choose two.)

Accepted Answer

Output encoding and filtering. Prompt injection attacks exploit the model's inability to distinguish between instructions and data. Input validation and sanitization can block malicious prompts before they reach the model, while output encoding and filtering can prevent sensitive data from being leaked even if an injection succeeds. These two controls provide defense in depth. Model quantization, temperature adjustments, and context window size do not address the root cause or impact of prompt injection.

Answer

Model quantization

Answer

Using a larger context window

Answer

Increasing the model's temperature

Question 3

When managing shadow AI, which strategy is most effective at the enterprise level?

Accepted Answer

Implementing AI-aware API discovery and CASB solutions.. Shadow AI — unsanctioned use of AI tools and APIs by employees — is best managed through AI-aware API discovery and Cloud Access Security Broker (CASB) solutions that detect, classify, and govern AI usage across the enterprise. These tools provide visibility into which AI services are being accessed, by whom, and with what data, enabling policy enforcement without blocking legitimate business activity. This balances security with productivity.

Answer

Disabling all network access to the internet.

Answer

Ignoring the usage until a breach occurs.

Answer

Sending disciplinary emails to all employees.

Question 4

An AI governance lead is drafting the organization's AI policy hierarchy. The board has approved a high-level AI principles statement. The lead now needs a document that translates those principles into mandatory, auditable requirements that apply to all AI systems, including those built by third parties on the organization's behalf. Which document type best fits this need?

Accepted Answer

An AI standard that specifies mandatory controls, roles, and evidence requirements mapped to each principle.. Policy hierarchies typically run from principles at the top, to standards that make requirements mandatory and auditable, to procedures that describe execution. The board-approved principles need a binding translation that applies across all AI systems, including third-party-built ones, and that specifies controls, roles, and evidence. A standard fills that layer and can be incorporated into third-party contracts, unlike procedures, position papers, or charters.

Answer

An AI position paper that explains the organization's philosophy and intended direction on AI ethics.

Answer

A set of AI procedures that describe step-by-step how individual teams perform model reviews.

Answer

An AI charter that establishes the governance committee's membership and meeting cadence.

Question 5

A global retailer is deploying an AI demand forecasting system that will automatically place supplier orders. The AI risk manager must establish controls that address the risk of automation bias, where human planners may defer to AI recommendations even when they are wrong. Which TWO controls BEST address this risk? (Choose two.)

Accepted Answer

Require planners to record an independent forecast before viewing the AI recommendation for a sample of orders.. Automation bias is best countered by controls that force independent human judgment and by information that helps humans calibrate their trust. Capturing a forecast before the AI output is shown breaks the anchoring cycle, while calibrated confidence scores tell planners when to be skeptical. Together they preserve meaningful human oversight, which is the control objective for consequential automated ordering decisions.

Answer

Increase the model's training data volume by tenfold to reduce the frequency of incorrect predictions.

Answer

Route all supplier orders through a fully automated approval workflow to eliminate inconsistent human decisions.

Answer

Grant planners full override authority without requiring justification for deviating from AI recommendations.

Question 6

What is a 'Side-Channel Attack' against an AI model?

Accepted Answer

An attack that exploits timing or power consumption patterns.. A side-channel attack infers secret information (such as model parameters or training data) by observing indirect physical or behavioral characteristics like timing, power consumption, electromagnetic emissions, or cache behavior. It does not attack the algorithm directly but exploits implementation leakage. Timing and power analysis are classic examples of this class.

Answer

An attack that uses a phishing email.

Answer

An attack that uses a VPN.

Answer

An attack that involves physical tampering.

Question 7

You are managing a PyTorch model and suspect data poisoning during the training phase. Which practice is the most effective technical control to verify training data integrity in an MLOps pipeline?

Accepted Answer

Implementing 'Data Lineage' with checksum validation. Data lineage with checksum validation is the most effective technical control because it creates a verifiable cryptographic chain of custody for training data. Checksums (e.g., SHA-256) detect any tampering or corruption of datasets between their source and the training job, while lineage tracks provenance, transformations, and access. Together they allow an MLOps pipeline to detect poisoned or substituted data before it influences model weights.

Answer

Increasing model quantization

Answer

Reducing the number of training epochs

Answer

Enabling 'Multi-Factor Authentication' on the training cluster

Question 8

Which tool is best for managing 'Secrets' (like database passwords or API keys) used by your AI training pipeline?

Accepted Answer

HashiCorp Vault. HashiCorp Vault is purpose-built for secrets management, offering dynamic secrets, encryption as a service, leasing and revocation, and detailed audit logs. It integrates with identity providers and cloud IAM to issue short-lived credentials for databases, API keys, and AI pipeline components, so no static secret ever lives in code or config. For an AI training pipeline that touches sensitive datasets and third-party model APIs, Vault's centralized policy engine and audit trail directly support governance and least-privilege requirements.

Answer

Hardcoded variables in the training script

Answer

A public Github repository

Answer

An Excel spreadsheet on a shared drive

Question 9

Which THREE technical controls effectively manage 'Third-Party AI Vendor' risks?

Accepted Answer

Independent verification of the vendor's model performance.. Option A is correct because independent verification of the vendor's model performance provides objective, third-party-validated evidence that the AI system meets accuracy, fairness, and reliability requirements, rather than relying solely on the vendor's self-reported benchmarks. Option B is correct because continuous monitoring of the vendor API for drift detects changes in model outputs, data distributions, or behavior over time (e.g., concept drift or data drift), enabling timely remediation of degraded or manipulated model performance. Option E is correct because mandatory security audits of the vendor's infrastructure assess the vendor's controls for data protection, access management, and resilience, which is essential for managing supply-chain risk in third-party AI services. Option C does not belong because asking a vendor to delete all internal logs is operationally unrealistic and would undermine auditability, incident response, and forensic traceability rather than manage risk. Option D does not belong because requiring employees to use the vendor's internal Wi-Fi is irrelevant to third-party AI vendor risk and could introduce additional security and privacy exposures.

Answer

Asking the vendor to delete all their internal logs.

Answer

Requiring all employees to use the vendor's internal Wi-Fi.

Question 10

An organization is deploying a machine learning model that makes automated decisions about loan applications. To comply with regulations, they need to ensure that the model's decisions can be explained to applicants and auditors. Which property of the model is most important to achieve this?

Accepted Answer

Interpretability. The scenario requires that the model's decisions be explained to applicants and auditors for regulatory compliance. Interpretability is the property that enables human understanding of how the model reaches its decisions. Scalability, latency, and accuracy are performance-related and do not address the need for explanation. Therefore, interpretability is the most important property in this context.

Answer

Accuracy

Answer

Latency

Answer

Scalability

Question 11

What does 'Model Explainability' (XAI) primarily support in governance?

Accepted Answer

Transparency, trust, and regulatory compliance.. Model Explainability (XAI) supports governance by making AI decision-making transparent and interpretable, which builds trust with stakeholders and satisfies regulatory requirements for explainability (e.g., GDPR Article 22, EU AI Act high-risk obligations). It enables auditors, regulators, and affected individuals to understand why a model produced a given output. Cost, hardware security, and training speed are unrelated to XAI's governance purpose.

Answer

Lower infrastructure costs.

Answer

Hardware security.

Answer

Increased training speed.

Question 12

An AI project is failing to meet its security requirements. Which governance process should be triggered to address this?

Accepted Answer

Risk acceptance or remediation plan initiation.. When an AI project fails to meet security requirements, the appropriate governance response is to initiate a formal risk acceptance or remediation plan. This ensures the risk is documented, evaluated against the organization's risk appetite, and either mitigated through a remediation roadmap or explicitly accepted by an accountable authority. This process maintains auditability and aligns with NIST AI RMF Manage and Govern functions.

Answer

Re-training the model without a review.

Answer

Immediate deletion of all data.

Answer

Ignoring the policy until the end of the year.

Question 13

A financial services firm has deployed a credit-scoring model, but the model owner cannot produce a defensible rationale for the weighting of each input feature, and the model card omits the intended use and known limitations. The CISO wants to align the AI system with ISACA's AI risk management expectations. Which action should be taken FIRST to address this gap?

Accepted Answer

Complete the model card and document feature rationale, intended use, and known limitations as part of the AI system inventory.. The scenario describes a governance and documentation deficiency: the model lacks a defensible feature rationale and omits intended use and known limitations. Completing the model card and documenting these elements restores traceability and accountability, which are foundational AI risk management activities. Other options address technical security testing, model redesign, or data protection, none of which close the documentation gap that prevents the business from understanding and accepting the model's residual risk.

Answer

Commission a full adversarial red-team exercise against the credit-scoring model before any documentation changes are made.

Answer

Immediately retrain the model with a different algorithm that offers inherently higher interpretability.

Answer

Encrypt the training dataset at rest and restrict access to the model's API endpoint.

Question 14

Which of the following is an example of an 'Evasion' attack on a computer vision model?

Accepted Answer

Adding adversarial noise to input images. An evasion attack on a computer vision model involves crafting inputs that are misclassified by the model while appearing normal to humans. Adding adversarial noise — small, often imperceptible perturbations to pixel values — is the canonical example: the modified image looks unchanged to a human but causes the model to output an incorrect label. This is a test-time attack that exploits the model's learned decision boundaries without altering the model itself.

Answer

Intercepting training data packets

Answer

Changing the model's learning rate

Answer

Brute forcing the administrator password

Question 15

A mid-sized insurance company is building its first AI risk inventory. The CISO asks the AI governance lead to define the scope of systems that must be registered in the inventory. Which approach best aligns with AI risk management principles?

Accepted Answer

Register all AI and machine learning systems that inform or make decisions affecting customers, employees, or material business operations.. An AI risk inventory should be scoped by impact, not by technology type or development origin. Registering every system that informs or makes decisions affecting customers, employees, or material operations ensures governance covers predictive and generative models alike, whether built or procured. This enables portfolio-level visibility, dependency analysis, and proportionate oversight. Narrow scoping by novelty, origin, or audit status creates blind spots.

Answer

Register only generative AI systems, since those present the newest and least understood risks.

Answer

Register only models developed in-house, since third-party models are the vendor's responsibility.

Answer

Register systems only after they have completed a full independent audit and received a fairness certification.