AAISM Practice Test 22 — 15 Questions

Question 1

A software company is rolling out an AI governance training program. The Chief Information Security Officer wants to ensure that all employees who build, deploy, or use AI systems understand the organization's expectations for responsible behavior, including prohibited uses. Which governance activity most directly establishes these expectations for the workforce?

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

Accepted Answer

Publishing an AI acceptable use policy that defines permitted, restricted, and prohibited AI behaviors. Expectations for responsible AI behavior are established through a published acceptable use policy that enumerates permitted, restricted, and prohibited uses. Training, monitoring, and bounty programs support the policy, but the policy itself is the instrument that defines expectations and provides the basis for consistent enforcement across the workforce.

Answer

Establishing a bug bounty program that rewards employees for reporting AI model vulnerabilities

Answer

Deploying an AI-powered anomaly detection system to monitor employee endpoint activity

Answer

Requiring annual security awareness training that includes a general module on emerging technologies

Question 2

An AI governance team is mapping controls for a credit-scoring model to the NIST AI Risk Management Framework. The team has completed the MAP function and is now working on MEASURE. Which activity is MOST appropriate for the MEASURE function?

Accepted Answer

Applying fairness metrics and robustness testing to quantify the model's discriminatory and adversarial risk.. The NIST AI RMF MEASURE function is about analyzing and assessing AI risks using metrics, methods, and tools. Applying fairness metrics and robustness testing quantifies discriminatory and adversarial risk, which is a direct MEASURE activity. Documenting context belongs to MAP, prioritizing resources belongs to MANAGE, and defining tolerance is a governance precursor.

Answer

Defining the organization's risk tolerance and appetite for AI systems.

Answer

Establishing a process to prioritize and allocate resources to identified AI risks.

Answer

Documenting the model's intended purpose, stakeholders, and deployment context.

Question 3

How does 'Continuous Monitoring' differ from 'Point-in-Time Auditing'?

Accepted Answer

Continuous monitoring provides real-time visibility into performance and threats.. Continuous monitoring provides real-time (or near-real-time) visibility into system performance, threats, and compliance posture, enabling immediate detection and response to anomalies. In contrast, point-in-time auditing captures a snapshot of controls and compliance at a specific moment, which can miss threats that emerge between audits. Continuous monitoring is a core principle of frameworks like NIST SP 800-137 and is essential for dynamic AI and cloud environments.

Answer

Auditing is only for small companies.

Answer

Monitoring is only for external systems.

Answer

Continuous monitoring is cheaper.

Question 4

Which TWO items must be documented in an 'AI Model Card'?

Accepted Answer

Known limitations and failure modes.. Option A is correct because an AI Model Card must document known limitations and failure modes, which are essential for transparency about where and how the model may fail or produce unreliable outputs. Option E is correct because the model's intended use cases are a core section of a Model Card, specifying the tasks and contexts the model was designed for and helping users avoid misuse. Option B is incorrect because a model's website color scheme is a cosmetic design detail unrelated to model documentation. Option C is incorrect because developers' social security numbers are sensitive personal data that have no place in a Model Card. Option D is incorrect because hardware purchase receipts are procurement records, not information about the model's capabilities, limitations, or intended use.

Answer

The color scheme of the model's website.

Answer

The social security numbers of all developers.

Answer

The hardware purchase receipts.

Question 5

Which mechanism best ensures 'Auditability' of AI decision-making?

Accepted Answer

Maintaining immutable logs of inputs, model versions, and outputs.. Auditability of AI decision-making requires a traceable record of what inputs were provided, which model version was used, and what outputs were produced. Maintaining immutable logs of these elements ensures that decisions can be reviewed, reproduced, and audited for compliance, bias, or errors. This is the foundational mechanism for AI auditability.

Answer

Deleting logs after 24 hours.

Answer

Using proprietary black-box models only.

Answer

Hardcoding all decisions.

Question 6

An enterprise AI governance committee is defining a policy for generative AI tools used by employees. The CISO wants to ensure that any AI-generated content that is published externally or used in decision-making is reviewed by a human before release. Which governance control should be formally documented to enforce this requirement across all business units?

Accepted Answer

A human-in-the-loop (HITL) review requirement in the AI acceptable use policy. A documented human-in-the-loop requirement in the AI acceptable use policy is the only option that creates a preventive, auditable governance control mandating human validation before AI-generated content is published or used in decisions. DLP, training, and audits are supporting measures but do not enforce the specific review gate the CISO requires.

Answer

An annual third-party audit of the AI governance program

Answer

A data loss prevention (DLP) rule that blocks uploads to public generative AI sites

Answer

A mandatory AI awareness training module for all employees

Question 7

Which THREE elements are essential components of an effective AI Incident Response Plan?

Accepted Answer

Procedures for model rollback or containment. Option A is correct because an effective AI Incident Response Plan must include procedures for model rollback or containment, such as reverting to a previous model version, disabling an API endpoint, or isolating the inference service, so that a misbehaving or compromised model can be quickly neutralized. Option D is correct because AI-specific incident severity levels are needed to classify events like model drift, data poisoning, prompt injection, or biased outputs, which do not map cleanly onto traditional IT severity scales and drive appropriate escalation and response timelines. Option E is correct because data scientists and AI model owners possess the domain knowledge of training data, model architecture, and expected behavior required to diagnose and remediate AI incidents, making their involvement essential to the response team. Option B does not belong because physical site evacuation protocols address physical safety emergencies and are not an AI-specific incident response component. Option C does not belong because mandatory weekly OS kernel patching is a routine vulnerability management control, not an essential element of an AI Incident Response Plan.

Answer

Physical site evacuation protocols

Answer

Mandatory weekly patching of all OS kernels

Question 8

A security manager is integrating AI governance into the existing NIST AI RMF framework. Which action best ensures alignment between AI security controls and business risk appetite?

Accepted Answer

Establishing a cross-functional AI steering committee to review and approve risk tolerance thresholds for AI use cases.. Aligning AI security controls with business risk appetite requires governance, not just technology. A cross-functional AI steering committee (C) brings together security, legal, data science, and business owners to define, review, and approve risk tolerance thresholds for AI use cases, ensuring controls map to what the business is actually willing to accept. This is the NIST AI RMF 'Govern' function in action, linking MAP, MEASURE, and MANAGE outcomes to organizational risk posture.

Answer

Deploying automated adversarial attack simulations against all LLM endpoints.

Answer

Implementing a strict zero-trust policy for all internal AI model training datasets.

Answer

Configuring AWS GuardDuty for Amazon Bedrock to monitor for unauthorized API calls.

Question 9

You are auditing an AI system for 'Model Explainability' (XAI). Why is XAI critical from a risk management perspective?

Accepted Answer

It provides transparency for accountability and bias detection.. Explainable AI (XAI) matters for risk management because it makes model decisions transparent and interpretable, which is essential for accountability (who is responsible and why a decision was made) and for detecting bias, discrimination, and unsafe behavior. Regulators and auditors increasingly require explanations for automated decisions affecting individuals, so XAI is a governance control rather than a performance optimization.

Answer

It is required to make the model run faster.

Answer

It encrypts the input data.

Answer

It enables models to train without labels.

Question 10

What is the primary risk of 'Over-reliance' on AI outputs?

Accepted Answer

Users may accept incorrect or harmful AI-generated information.. Over-reliance on AI outputs occurs when users treat AI-generated content as authoritative without independent verification, leading them to accept incorrect, biased, or harmful information. Because AI models can hallucinate or produce plausible-sounding but false outputs, unchecked trust creates direct business, safety, and compliance risk. The primary risk is therefore the acceptance of bad outputs, not any performance or infrastructure issue.

Answer

The model will perform too well.

Answer

The server will overheat.

Answer

The model runs out of memory.

Question 11

A hospital's AI governance lead is classifying an AI triage tool that prioritizes emergency department patients. The tool was trained on five years of historical admission data. A risk assessment must determine the inherent risk tier before any controls are applied. Which factor most directly drives the inherent risk classification of this system?

Accepted Answer

The potential for direct physical harm to patients if the model prioritizes incorrectly.. Inherent risk classification is driven by the severity of consequences to affected parties before any controls are applied. A triage model that can misorder critically ill patients creates potential for direct physical harm, which places it in the highest tier. Dataset size, model architecture, and hosting region are relevant to control design and model quality but do not determine the inherent risk tier.

Answer

Whether the model uses a transformer or gradient-boosted tree architecture.

Answer

The cloud region where the model's inference endpoints are hosted.

Answer

The volume of training records used to build the model.

Question 12

In the context of 'Adversarial Machine Learning', what is a 'Blind Spot' attack?

Accepted Answer

Exploiting low-confidence regions to force incorrect predictions. A Blind Spot attack in adversarial machine learning exploits regions of the input space where the model has low confidence or sparse training coverage, crafting inputs that force incorrect predictions. By targeting these under-trained regions, the attacker can cause misclassification without needing large perturbations, because the model has not learned robust decision boundaries there. This is distinct from evasion attacks that rely on gradient signals in high-confidence regions.

Answer

Bypassing authentication on the server

Answer

Increasing the training time of the model

Answer

Deleting the model's weights from memory

Question 13

Which metric is most appropriate for reporting the 'AI Security Posture' to the board?

Accepted Answer

Percentage of AI systems with completed risk assessments.. Board-level reporting on AI security posture should reflect governance effectiveness and risk coverage, not operational throughput. The percentage of AI systems with completed risk assessments directly measures how much of the AI estate has been evaluated against security and compliance criteria. It is an actionable, trendable governance metric that boards can use to hold management accountable.

Answer

Number of GPU hours consumed.

Answer

Average model inference accuracy.

Answer

Total number of model versions created.

Question 14

A financial services firm has deployed an AI model that flags suspicious transactions. After six months, monitoring shows the model's false-positive rate has drifted from 4% to 11%, generating excessive analyst workload. The data science team confirms the model has not been retrained and the feature pipeline is unchanged. Which risk management action should the AI governance lead prioritize?

Accepted Answer

Investigate the cause of drift by comparing current input data distributions against the training baseline, then decide on remediation.. Performance drift without retraining or pipeline changes points to a shift in data distributions or in the input-to-outcome relationship. The governance lead should first diagnose whether the cause is data drift or concept drift by comparing current inputs to the training baseline. That diagnosis determines whether retraining, recalibration, or upstream data remediation is appropriate. Retraining, threshold tuning, and deferral all act without understanding the cause.

Answer

Document the drift in the risk register and schedule a review at the next quarterly governance meeting.

Answer

Immediately retrain the model on the most recent transaction data and redeploy.

Answer

Raise the model's decision threshold so fewer transactions are flagged and analyst workload returns to normal.

Question 15

A financial institution is deploying a large language model (LLM) for customer support. The security team wants to ensure that the model does not inadvertently reveal sensitive information from its training data. They decide to implement a technique that adds noise to the training process to provide a formal privacy guarantee. Which technique should they use?

Accepted Answer

Differential privacy. Differential privacy is the only technique listed that provides a formal privacy guarantee by adding noise during training, making it difficult for an attacker to determine if a specific record was used. Homomorphic encryption, secure multi-party computation, and federated learning address different privacy aspects but do not directly prevent a model from revealing training data.

Answer

Homomorphic encryption

Answer

Federated learning

Answer

Secure multi-party computation