cert
cert-1
cert-2

Pass Isaca AAIA Exam in First Attempt Guaranteed!

Get 100% Latest Exam Questions, Accurate & Verified Answers to Pass the Actual Exam!
30 Days Free Updates, Instant Download!

cert-5
cert-6
AAIA Exam - Verified By Experts
AAIA Premium Bundle
$14.99

AAIA Premium Bundle

$74.99
$89.98
  • Premium File 556 Questions & Answers. Last update: Nov 08, 2026
  • Training Course 110 Video Lectures
 
$89.98
$74.99
accept 22 downloads in last 7 days
block-screenshots
Purchase Isaca AAIA Exam Training Products Individually
Isaca AAIA Premium File
Premium File 556 Q&A
$76.99$69.99
Isaca AAIA Video Training Course
Training Course 110 Lectures
$21.99 $19.99
Isaca AAIA Practice Test 79

Question 780

An independent AI auditor reviews this case: A legal assistant detects one jailbreak phrase but fails paraphrased attempts after a prompt update. Which TWO complementary elements should the auditor seek: an effective corrective control and the most probative independent evidence? Select TWO answers.

  1. Call one successful refusal to an obvious unsafe prompt a comprehensive security red-team assessment.
  2. Inspect threat model, attack corpus diversity, tool permissions, exploit transcripts and remediation evidence.
  3. Test citation support, omissions, contradiction handling and safe abstention using independently labeled examples.
  4. Conduct scoped adversarial testing across inputs, retrieved context, tools, identities and post-change regressions.
  5. Inspect stratified confusion matrices, sample sizes, relevant context, confidence intervals and exception responses.

Explanation

Correct Answers

B:

The reason it is necessary is: AI misuse testing requires plausible attacker-controlled surfaces and repeatable negative cases. For this situation: A legal assistant detects one jailbreak phrase but fails paraphrased attempts after a prompt update.

D:

The reason it is necessary is: AI misuse testing requires plausible attacker-controlled surfaces and repeatable negative cases. This response fits the risk associated with AI adversarial red-team scope and regression: Conduct scoped adversarial testing across inputs, retrieved context, tools, identities and post-change regressions.

Incorrect Answers

A:

It misses the key control principle: AI misuse testing requires plausible attacker-controlled surfaces and repeatable negative cases.

C:

This option concerns Grounded generative answer quality evaluation, which is a related but different AI governance concern. In this case: A legal assistant detects one jailbreak phrase but fails paraphrased attempts after a prompt update.

E:

This option concerns Subgroup fairness performance validation, which is a related but different AI governance concern. For this situation: A legal assistant detects one jailbreak phrase but fails paraphrased attempts after a prompt update.

Question 781

A governance and assurance review uncovers the following: A virtual assistant meets average response targets but peaks cause 40-second tail latency during sales events. What is the MOST appropriate risk-based audit recommendation that addresses the root control weakness? Select ONE best answer.

  1. Approve production based solely on one average latency measurement in a quiet lab.
  2. Evaluate behavior under credible distribution shifts and reject or safely defer unsupported inputs.
  3. Load-test realistic context sizes, concurrency, cost limits, percentiles and fallback behavior.
  4. Test extractable memorization and membership inference under authorized red-team conditions and apply data minimization.

Explanation

Correct Answer

C:

The acceptable verified outcome is: Realistic peak workloads meet approved quality, latency and cost limits with graceful degradation. This is the proportionate control response to AI service latency cost and capacity testing: Load-test realistic context sizes, concurrency, cost limits, percentiles and fallback behavior.

Incorrect Answers

A:

This attractive shortcut does not resolve the AI service latency cost and capacity testing weakness: Approve production based solely on one average latency measurement in a quiet lab.

B:

This option concerns Out-of-distribution robustness and shift testing, which is a related but different AI governance concern. It does not establish control effectiveness for AI service latency cost and capacity testing.

D:

This option concerns Membership inference and training-data memorization tests, which is a related but different AI governance concern. It does not establish control effectiveness for AI service latency cost and capacity testing.

Question 782

A risk-based audit engagement examines this operating condition: A virtual assistant meets average response targets but peaks cause 40-second tail latency during sales events. Which principle BEST explains why the observed practice is a material governance or control concern? Select ONE best answer.

  1. Approve production based solely on one average latency measurement in a quiet lab.
  2. Fluency and coherence are not evidence of factual correctness or fidelity to source material.
  3. Averages hide tail failures and resource exhaustion; service acceptability includes efficiency and availability.
  4. Model validation must reflect foreseeable deployment conditions, not only random examples from one source.

Explanation

Correct Answer

C:

The corrective action is: Load-test realistic context sizes, concurrency, cost limits, percentiles and fallback behavior. The acceptable verified outcome is: Realistic peak workloads meet approved quality, latency and cost limits with graceful degradation.

Incorrect Answers

A:

This attractive shortcut does not resolve the AI service latency cost and capacity testing weakness: Approve production based solely on one average latency measurement in a quiet lab. For this situation: A virtual assistant meets average response targets but peaks cause 40-second tail latency during sales events.

B:

This option concerns Grounded generative answer quality evaluation, which is a related but different AI governance concern. It does not establish control effectiveness for AI service latency cost and capacity testing.

D:

This option concerns Out-of-distribution robustness and shift testing, which is a related but different AI governance concern. For this situation: A virtual assistant meets average response targets but peaks cause 40-second tail latency during sales events.

Question 783

An audit committee asks for a defensible assessment of this situation: A virtual assistant meets average response targets but peaks cause 40-second tail latency during sales events. What evidence should the auditor examine FIRST to substantiate or challenge the control owner’s claim? Select ONE best answer.

  1. Inspect p95/p99 response, token counts, queue depth, error budgets, concurrency and billing ceilings.
  2. Inspect training/evaluation overlap, duplicate hashes, benchmark dates, partition strategy and blind results.
  3. Inspect dataset snapshots, feature pipeline commits, record counts, model registry links and training execution manifests.
  4. Approve production based solely on one average latency measurement in a quiet lab.

Explanation

Correct Answer

A:

The corrective action is: Load-test realistic context sizes, concurrency, cost limits, percentiles and fallback behavior. In this case: A virtual assistant meets average response targets but peaks cause 40-second tail latency during sales events.

Incorrect Answers

B:

This option concerns Training and evaluation benchmark contamination, which is a related but different AI governance concern. It does not establish control effectiveness for AI service latency cost and capacity testing.

C:

This option concerns AI training dataset lineage and reproducibility, which is a related but different AI governance concern. It does not establish control effectiveness for AI service latency cost and capacity testing.

D:

This attractive shortcut does not resolve the AI service latency cost and capacity testing weakness: Approve production based solely on one average latency measurement in a quiet lab. It misses the key control principle: Averages hide tail failures and resource exhaustion; service acceptability includes efficiency and availability.

Question 784

During an AI controls assessment, the team learns that a virtual assistant meets average response targets but peaks cause 40-second tail latency during sales events. After management addresses the issue, which result MOST convincingly demonstrates effective remediation? Select ONE best answer.

  1. Realistic peak workloads meet approved quality, latency and cost limits with graceful degradation.
  2. Approve production based solely on one average latency measurement in a quiet lab.
  3. Model performance remains within approved bounds or unsafe inputs reliably route to fallback.
  4. Critical attack paths are blocked by tested technical controls and regressions catch recurrence.

Explanation

Correct Answer

A:

The corrective action is: Load-test realistic context sizes, concurrency, cost limits, percentiles and fallback behavior. This is an observable validation result for AI service latency cost and capacity testing, not just management assurance: Realistic peak workloads meet approved quality, latency and cost limits with graceful degradation.

Incorrect Answers

B:

This attractive shortcut does not resolve the AI service latency cost and capacity testing weakness: Approve production based solely on one average latency measurement in a quiet lab. The auditor should instead examine: Inspect p95/p99 response, token counts, queue depth, error budgets, concurrency and billing ceilings.

C:

This option concerns Out-of-distribution robustness and shift testing, which is a related but different AI governance concern. The decisive principle here is: Averages hide tail failures and resource exhaustion; service acceptability includes efficiency and availability.

D:

This option concerns AI adversarial red-team scope and regression, which is a related but different AI governance concern. It does not establish control effectiveness for AI service latency cost and capacity testing.

Question 785

An independent AI auditor reviews this case: An inference model has acceptable quality yet batch jobs consume GPU memory and trigger uncontrolled cost overrun. What is the MOST appropriate risk-based audit recommendation that addresses the root control weakness? Select ONE best answer.

  1. Use independent held-out evaluations with leakage checks and repeatable challenge sets not tuned into training.
  2. Conduct scoped adversarial testing across inputs, retrieved context, tools, identities and post-change regressions.
  3. Load-test realistic context sizes, concurrency, cost limits, percentiles and fallback behavior.
  4. Approve production based solely on one average latency measurement in a quiet lab.

Explanation

Correct Answer

C:

The acceptable verified outcome is: Realistic peak workloads meet approved quality, latency and cost limits with graceful degradation. The reason it is necessary is: Averages hide tail failures and resource exhaustion; service acceptability includes efficiency and availability.

Incorrect Answers

A:

This option concerns Training and evaluation benchmark contamination, which is a related but different AI governance concern. The decisive principle here is: Averages hide tail failures and resource exhaustion; service acceptability includes efficiency and availability.

B:

This option concerns AI adversarial red-team scope and regression, which is a related but different AI governance concern. The decisive principle here is: Averages hide tail failures and resource exhaustion; service acceptability includes efficiency and availability.

D:

This attractive shortcut does not resolve the AI service latency cost and capacity testing weakness: Approve production based solely on one average latency measurement in a quiet lab. Load-test realistic context sizes, concurrency, cost limits, percentiles and fallback behavior.

Question 786

A governance and assurance review uncovers the following: An inference model has acceptable quality yet batch jobs consume GPU memory and trigger uncontrolled cost overrun. Which principle BEST explains why the observed practice is a material governance or control concern? Select ONE best answer.

  1. Averages hide tail failures and resource exhaustion; service acceptability includes efficiency and availability.
  2. Approve production based solely on one average latency measurement in a quiet lab.
  3. Aggregate accuracy can conceal materially unequal errors or harms for specific populations.
  4. Fluency and coherence are not evidence of factual correctness or fidelity to source material.

Explanation

Correct Answer

A:

The corrective action is: Load-test realistic context sizes, concurrency, cost limits, percentiles and fallback behavior. This is the applicable AI audit principle for AI service latency cost and capacity testing: Averages hide tail failures and resource exhaustion; service acceptability includes efficiency and availability.

Incorrect Answers

B:

This attractive shortcut does not resolve the AI service latency cost and capacity testing weakness: Approve production based solely on one average latency measurement in a quiet lab. For this situation: An inference model has acceptable quality yet batch jobs consume GPU memory and trigger uncontrolled cost overrun.

C:

This option concerns Subgroup fairness performance validation, which is a related but different AI governance concern. It does not establish control effectiveness for AI service latency cost and capacity testing.

D:

This option concerns Grounded generative answer quality evaluation, which is a related but different AI governance concern. The decisive principle here is: Averages hide tail failures and resource exhaustion; service acceptability includes efficiency and availability.

Question 787

A risk-based audit engagement examines this operating condition: An inference model has acceptable quality yet batch jobs consume GPU memory and trigger uncontrolled cost overrun. What evidence should the auditor examine FIRST to substantiate or challenge the control owner’s claim? Select ONE best answer.

  1. Inspect dataset snapshots, feature pipeline commits, record counts, model registry links and training execution manifests.
  2. Approve production based solely on one average latency measurement in a quiet lab.
  3. Inspect source passage matches, structured evaluation rubrics, omission rates, unsupported-claim samples and reviewer disagreements.
  4. Inspect p95/p99 response, token counts, queue depth, error budgets, concurrency and billing ceilings.

Explanation

Correct Answer

D:

The corrective action is: Load-test realistic context sizes, concurrency, cost limits, percentiles and fallback behavior. This evidence directly tests the asserted control for AI service latency cost and capacity testing: Inspect p95/p99 response, token counts, queue depth, error budgets, concurrency and billing ceilings.

Incorrect Answers

A:

This option concerns AI training dataset lineage and reproducibility, which is a related but different AI governance concern. The decisive principle here is: Averages hide tail failures and resource exhaustion; service acceptability includes efficiency and availability.

B:

This attractive shortcut does not resolve the AI service latency cost and capacity testing weakness: Approve production based solely on one average latency measurement in a quiet lab. Effective closure requires: Realistic peak workloads meet approved quality, latency and cost limits with graceful degradation.

C:

This option concerns Grounded generative answer quality evaluation, which is a related but different AI governance concern. For this situation: An inference model has acceptable quality yet batch jobs consume GPU memory and trigger uncontrolled cost overrun.

Question 788

An audit committee asks for a defensible assessment of this situation: An inference model has acceptable quality yet batch jobs consume GPU memory and trigger uncontrolled cost overrun. After management addresses the issue, which result MOST convincingly demonstrates effective remediation? Select ONE best answer.

  1. An independent rerun can identify the exact source data and material transformation decisions for a sampled model version.
  2. Approve production based solely on one average latency measurement in a quiet lab.
  3. Critical attack paths are blocked by tested technical controls and regressions catch recurrence.
  4. Realistic peak workloads meet approved quality, latency and cost limits with graceful degradation.

Explanation

Correct Answer

D:

The corrective action is: Load-test realistic context sizes, concurrency, cost limits, percentiles and fallback behavior. The reason it is necessary is: Averages hide tail failures and resource exhaustion; service acceptability includes efficiency and availability.

Incorrect Answers

A:

This option concerns AI training dataset lineage and reproducibility, which is a related but different AI governance concern. For this situation: An inference model has acceptable quality yet batch jobs consume GPU memory and trigger uncontrolled cost overrun.

B:

For this situation: An inference model has acceptable quality yet batch jobs consume GPU memory and trigger uncontrolled cost overrun.

C:

This option concerns AI adversarial red-team scope and regression, which is a related but different AI governance concern. For this situation: An inference model has acceptable quality yet batch jobs consume GPU memory and trigger uncontrolled cost overrun.

Question 789

During an AI controls assessment, the team learns that a real-time risk service tests throughput only with short prompts although production uses long retrieved contexts. What is the MOST appropriate risk-based audit recommendation that addresses the root control weakness? Select ONE best answer.

  1. Version and trace training, validation and inference datasets with accountable lineage, hashes and reproducible pipeline metadata.
  2. Approve production based solely on one average latency measurement in a quiet lab.
  3. Load-test realistic context sizes, concurrency, cost limits, percentiles and fallback behavior.
  4. Evaluate justified subgroup error and outcome metrics with adequate samples and documented mitigation thresholds.

Explanation

Correct Answer

C:

The acceptable verified outcome is: Realistic peak workloads meet approved quality, latency and cost limits with graceful degradation. In this case: A real-time risk service tests throughput only with short prompts although production uses long retrieved contexts.

Incorrect Answers

A:

This option concerns AI training dataset lineage and reproducibility, which is a related but different AI governance concern. In this case: A real-time risk service tests throughput only with short prompts although production uses long retrieved contexts.

B:

This attractive shortcut does not resolve the AI service latency cost and capacity testing weakness: Approve production based solely on one average latency measurement in a quiet lab. In this case: A real-time risk service tests throughput only with short prompts although production uses long retrieved contexts.

D:

This option concerns Subgroup fairness performance validation, which is a related but different AI governance concern. In this case: A real-time risk service tests throughput only with short prompts although production uses long retrieved contexts.

Question 790

An independent AI auditor reviews this case: A real-time risk service tests throughput only with short prompts although production uses long retrieved contexts. Which TWO complementary elements should the auditor seek: an effective corrective control and the most probative independent evidence? Select TWO answers.

  1. Load-test realistic context sizes, concurrency, cost limits, percentiles and fallback behavior.
  2. Approve production based solely on one average latency measurement in a quiet lab.
  3. Inspect controlled probe outputs, duplication controls, privacy evaluations, training-source permissions and mitigation.
  4. Use independent held-out evaluations with leakage checks and repeatable challenge sets not tuned into training.
  5. Inspect p95/p99 response, token counts, queue depth, error budgets, concurrency and billing ceilings.

Explanation

Correct Answers

A:

The acceptable verified outcome is: Realistic peak workloads meet approved quality, latency and cost limits with graceful degradation. For this situation: A real-time risk service tests throughput only with short prompts although production uses long retrieved contexts.

E:

The corrective action is: Load-test realistic context sizes, concurrency, cost limits, percentiles and fallback behavior. In this case: A real-time risk service tests throughput only with short prompts although production uses long retrieved contexts.

Incorrect Answers

B:

This attractive shortcut does not resolve the AI service latency cost and capacity testing weakness: Approve production based solely on one average latency measurement in a quiet lab. For this situation: A real-time risk service tests throughput only with short prompts although production uses long retrieved contexts.

C:

This option concerns Membership inference and training-data memorization tests, which is a related but different AI governance concern. In this case: A real-time risk service tests throughput only with short prompts although production uses long retrieved contexts.

D:

This option concerns Training and evaluation benchmark contamination, which is a related but different AI governance concern. In this case: A real-time risk service tests throughput only with short prompts although production uses long retrieved contexts.

ISACA Advanced in AI Audit (AAIA) Practice Tests