Isaca AAIA Practice Test 59
Question 580
An independent AI auditor reviews this case: A customer-support assistant's system prompt is edited directly in production after one negative feedback report. Which TWO complementary elements should the auditor seek: an effective corrective control and the most probative independent evidence? Select TWO answers.
- Inspect calibration, confusion matrices, missing-input behavior, escalation queue and approval history.
- Version prompt templates, permission boundaries and evaluation suites with staged approvals and rollback.
- Log emergency AI configuration changes, require proportionate approval, expiry and post-change validation.
- Allow any content editor to hot-swap production prompts and regard that as nontechnical copy editing.
- Inspect prompt history, version diffs, tested attack cases, tool allowlists and release tickets.
Explanation
Correct Answers
B:The reason it is necessary is: Prompts are executable behavioral configuration and can change output and risk without code deployment. This response fits the risk associated with Prompt template governance and injection-resilience changes: Version prompt templates, permission boundaries and evaluation suites with staged approvals and rollback.
E:The reason it is necessary is: Prompts are executable behavioral configuration and can change output and risk without code deployment. For this situation: A customer-support assistant's system prompt is edited directly in production after one negative feedback report.
Incorrect Answers
A:This option concerns Threshold governance and safety fallbacks, which is a related but different AI governance concern. The decisive principle here is: Prompts are executable behavioral configuration and can change output and risk without code deployment.
C:This option concerns AI configuration and emergency change traceability, which is a related but different AI governance concern. The decisive principle here is: Prompts are executable behavioral configuration and can change output and risk without code deployment.
D:This attractive shortcut does not resolve the Prompt template governance and injection-resilience changes weakness: Allow any content editor to hot-swap production prompts and regard that as nontechnical copy editing. Effective closure requires: Authorized prompt versions perform as specified and fail closed on tested malicious context.
Question 581
A governance and assurance review uncovers the following: A prompt change permits unauthorized tool invocations but is not linked to any version or test results. What is the MOST appropriate risk-based audit recommendation that addresses the root control weakness? Select ONE best answer.
- Classify model, data, prompt and feature changes by potential behavior impact and require appropriate regression testing.
- Allow any content editor to hot-swap production prompts and regard that as nontechnical copy editing.
- Version prompt templates, permission boundaries and evaluation suites with staged approvals and rollback.
- Control schema, source, feature and data refreshes using integrity and semantic compatibility tests.
Explanation
Correct Answer
C:The reason it is necessary is: Prompts are executable behavioral configuration and can change output and risk without code deployment. In this case: A prompt change permits unauthorized tool invocations but is not linked to any version or test results.
Incorrect Answers
A:This option concerns Model version change impact classification, which is a related but different AI governance concern. The decisive principle here is: Prompts are executable behavioral configuration and can change output and risk without code deployment.
B:This attractive shortcut does not resolve the Prompt template governance and injection-resilience changes weakness: Allow any content editor to hot-swap production prompts and regard that as nontechnical copy editing. In this case: A prompt change permits unauthorized tool invocations but is not linked to any version or test results.
D:This option concerns AI dataset refresh and feature pipeline approvals, which is a related but different AI governance concern. It does not establish control effectiveness for Prompt template governance and injection-resilience changes.
Question 582
A risk-based audit engagement examines this operating condition: A prompt change permits unauthorized tool invocations but is not linked to any version or test results. Which principle BEST explains why the observed practice is a material governance or control concern? Select ONE best answer.
- The same model weights can behave differently when feature definitions or data feeds change.
- Allow any content editor to hot-swap production prompts and regard that as nontechnical copy editing.
- AI safety depends on configuration state as well as model artifacts; emergency bypasses need disciplined control.
- Prompts are executable behavioral configuration and can change output and risk without code deployment.
Explanation
Correct Answer
D:Independent audit evidence includes: Inspect prompt history, version diffs, tested attack cases, tool allowlists and release tickets. In this case: A prompt change permits unauthorized tool invocations but is not linked to any version or test results.
Incorrect Answers
A:This option concerns AI dataset refresh and feature pipeline approvals, which is a related but different AI governance concern. For this situation: A prompt change permits unauthorized tool invocations but is not linked to any version or test results.
B:This attractive shortcut does not resolve the Prompt template governance and injection-resilience changes weakness: Allow any content editor to hot-swap production prompts and regard that as nontechnical copy editing. For this situation: A prompt change permits unauthorized tool invocations but is not linked to any version or test results.
C:This option concerns AI configuration and emergency change traceability, which is a related but different AI governance concern. In this case: A prompt change permits unauthorized tool invocations but is not linked to any version or test results.
Question 583
An audit committee asks for a defensible assessment of this situation: A prompt change permits unauthorized tool invocations but is not linked to any version or test results. What evidence should the auditor examine FIRST to substantiate or challenge the control owner’s claim? Select ONE best answer.
- Allow any content editor to hot-swap production prompts and regard that as nontechnical copy editing.
- Inspect config state histories, privileges, break-glass logs, time bounds and post-incident review.
- Inspect prompt history, version diffs, tested attack cases, tool allowlists and release tickets.
- Inspect source contracts, schema diffs, baseline distributions, feature tests and observed scoring changes.
Explanation
Correct Answer
C:The reason it is necessary is: Prompts are executable behavioral configuration and can change output and risk without code deployment. The corrective action is: Version prompt templates, permission boundaries and evaluation suites with staged approvals and rollback.
Incorrect Answers
A:This attractive shortcut does not resolve the Prompt template governance and injection-resilience changes weakness: Allow any content editor to hot-swap production prompts and regard that as nontechnical copy editing. Version prompt templates, permission boundaries and evaluation suites with staged approvals and rollback.
B:This option concerns AI configuration and emergency change traceability, which is a related but different AI governance concern. For this situation: A prompt change permits unauthorized tool invocations but is not linked to any version or test results.
D:This option concerns AI dataset refresh and feature pipeline approvals, which is a related but different AI governance concern. The decisive principle here is: Prompts are executable behavioral configuration and can change output and risk without code deployment.
Question 584
During an AI controls assessment, the team learns that a prompt change permits unauthorized tool invocations but is not linked to any version or test results. After management addresses the issue, which result MOST convincingly demonstrates effective remediation? Select ONE best answer.
- Allow any content editor to hot-swap production prompts and regard that as nontechnical copy editing.
- Authorized prompt versions perform as specified and fail closed on tested malicious context.
- Material external changes are assessed before or promptly after adoption with controlled escalation.
- Emergency deviations expire or receive accountable renewal and critical controls are verified restored.
Explanation
Correct Answer
B:The reason it is necessary is: Prompts are executable behavioral configuration and can change output and risk without code deployment. This is an observable validation result for Prompt template governance and injection-resilience changes, not just management assurance: Authorized prompt versions perform as specified and fail closed on tested malicious context.
Incorrect Answers
A:For this situation: A prompt change permits unauthorized tool invocations but is not linked to any version or test results.
C:This option concerns External foundation-model update change notices, which is a related but different AI governance concern. The decisive principle here is: Prompts are executable behavioral configuration and can change output and risk without code deployment.
D:This option concerns AI configuration and emergency change traceability, which is a related but different AI governance concern. The audit must corroborate: Inspect prompt history, version diffs, tested attack cases, tool allowlists and release tickets.
Question 585
An independent AI auditor reviews this case: A legal assistant produces different citation quality after an untracked retrieval instruction update. What is the MOST appropriate risk-based audit recommendation that addresses the root control weakness? Select ONE best answer.
- Log emergency AI configuration changes, require proportionate approval, expiry and post-change validation.
- Version prompt templates, permission boundaries and evaluation suites with staged approvals and rollback.
- Classify model, data, prompt and feature changes by potential behavior impact and require appropriate regression testing.
- Allow any content editor to hot-swap production prompts and regard that as nontechnical copy editing.
Explanation
Correct Answer
B:The reason it is necessary is: Prompts are executable behavioral configuration and can change output and risk without code deployment. Independent audit evidence includes: Inspect prompt history, version diffs, tested attack cases, tool allowlists and release tickets.
Incorrect Answers
A:This option concerns AI configuration and emergency change traceability, which is a related but different AI governance concern. In this case: A legal assistant produces different citation quality after an untracked retrieval instruction update.
C:This option concerns Model version change impact classification, which is a related but different AI governance concern. The audit must corroborate: Inspect prompt history, version diffs, tested attack cases, tool allowlists and release tickets.
D:This attractive shortcut does not resolve the Prompt template governance and injection-resilience changes weakness: Allow any content editor to hot-swap production prompts and regard that as nontechnical copy editing. In this case: A legal assistant produces different citation quality after an untracked retrieval instruction update.
Question 586
A governance and assurance review uncovers the following: A legal assistant produces different citation quality after an untracked retrieval instruction update. Which principle BEST explains why the observed practice is a material governance or control concern? Select ONE best answer.
- Allow any content editor to hot-swap production prompts and regard that as nontechnical copy editing.
- The same model weights can behave differently when feature definitions or data feeds change.
- Prompts are executable behavioral configuration and can change output and risk without code deployment.
- Gradual deployment reduces blast radius only if meaningful indicators and rollback decisions are predefined.
Explanation
Correct Answer
C:Independent audit evidence includes: Inspect prompt history, version diffs, tested attack cases, tool allowlists and release tickets. The corrective action is: Version prompt templates, permission boundaries and evaluation suites with staged approvals and rollback.
Incorrect Answers
A:This attractive shortcut does not resolve the Prompt template governance and injection-resilience changes weakness: Allow any content editor to hot-swap production prompts and regard that as nontechnical copy editing. For this situation: A legal assistant produces different citation quality after an untracked retrieval instruction update.
B:This option concerns AI dataset refresh and feature pipeline approvals, which is a related but different AI governance concern. The audit must corroborate: Inspect prompt history, version diffs, tested attack cases, tool allowlists and release tickets.
D:This option concerns Canary releases and model rollback thresholds, which is a related but different AI governance concern. The decisive principle here is: Prompts are executable behavioral configuration and can change output and risk without code deployment.
Question 587
A risk-based audit engagement examines this operating condition: A legal assistant produces different citation quality after an untracked retrieval instruction update. What evidence should the auditor examine FIRST to substantiate or challenge the control owner’s claim? Select ONE best answer.
- Inspect source contracts, schema diffs, baseline distributions, feature tests and observed scoring changes.
- Inspect tool-call traces, retrieval source trust, instruction hierarchy tests and denied action logs.
- Inspect prompt history, version diffs, tested attack cases, tool allowlists and release tickets.
- Allow any content editor to hot-swap production prompts and regard that as nontechnical copy editing.
Explanation
Correct Answer
C:The reason it is necessary is: Prompts are executable behavioral configuration and can change output and risk without code deployment. The acceptable verified outcome is: Authorized prompt versions perform as specified and fail closed on tested malicious context.
Incorrect Answers
A:This option concerns AI dataset refresh and feature pipeline approvals, which is a related but different AI governance concern. For this situation: A legal assistant produces different citation quality after an untracked retrieval instruction update.
B:This option concerns Indirect prompt injection through retrieved content, which is a related but different AI governance concern. The decisive principle here is: Prompts are executable behavioral configuration and can change output and risk without code deployment.
D:It misses the key control principle: Prompts are executable behavioral configuration and can change output and risk without code deployment.
Question 588
An audit committee asks for a defensible assessment of this situation: A legal assistant produces different citation quality after an untracked retrieval instruction update. After management addresses the issue, which result MOST convincingly demonstrates effective remediation? Select ONE best answer.
- Authorized prompt versions perform as specified and fail closed on tested malicious context.
- Material external changes are assessed before or promptly after adoption with controlled escalation.
- Malicious document instructions cannot cause prohibited tool actions or confidential-data release.
- Allow any content editor to hot-swap production prompts and regard that as nontechnical copy editing.
Explanation
Correct Answer
A:The reason it is necessary is: Prompts are executable behavioral configuration and can change output and risk without code deployment. In this case: A legal assistant produces different citation quality after an untracked retrieval instruction update.
Incorrect Answers
B:This option concerns External foundation-model update change notices, which is a related but different AI governance concern. The audit must corroborate: Inspect prompt history, version diffs, tested attack cases, tool allowlists and release tickets.
C:This option concerns Indirect prompt injection through retrieved content, which is a related but different AI governance concern. For this situation: A legal assistant produces different citation quality after an untracked retrieval instruction update.
D:For this situation: A legal assistant produces different citation quality after an untracked retrieval instruction update.
Question 589
During an AI controls assessment, the team learns that a bank deploys a new classifier to all customers after testing only one branch and later discovers disparate denials. What is the MOST appropriate risk-based audit recommendation that addresses the root control weakness? Select ONE best answer.
- Require contractual update notifications, version observability, revalidation and compensating fallback for material supplier changes.
- Call a deployment canary controlled when traffic fraction is small but no adverse-impact trigger is monitored.
- Use risk-aware canary populations, guardrail metrics, stop criteria and tested rollback or kill switch.
- Classify model, data, prompt and feature changes by potential behavior impact and require appropriate regression testing.
Explanation
Correct Answer
C:The reason it is necessary is: Gradual deployment reduces blast radius only if meaningful indicators and rollback decisions are predefined. The acceptable verified outcome is: Unacceptable changes are contained promptly and validated prior behavior is restored.
Incorrect Answers
A:This option concerns External foundation-model update change notices, which is a related but different AI governance concern. It does not establish control effectiveness for Canary releases and model rollback thresholds.
B:This attractive shortcut does not resolve the Canary releases and model rollback thresholds weakness: Call a deployment canary controlled when traffic fraction is small but no adverse-impact trigger is monitored.
D:This option concerns Model version change impact classification, which is a related but different AI governance concern. It does not establish control effectiveness for Canary releases and model rollback thresholds.
Question 590
An independent AI auditor reviews this case: A bank deploys a new classifier to all customers after testing only one branch and later discovers disparate denials. Which TWO complementary elements should the auditor seek: an effective corrective control and the most probative independent evidence? Select TWO answers.
- Call a deployment canary controlled when traffic fraction is small but no adverse-impact trigger is monitored.
- Use risk-aware canary populations, guardrail metrics, stop criteria and tested rollback or kill switch.
- Quarantine suspicious inputs and models, preserve provenance, revert to validated artifacts and investigate contamination.
- Inspect canary assignment, segment KPIs, error thresholds, baseline model artifacts and rollback drill.
- Compare before-after distributions, update artifacts, critical use cases, approval rules and rollback records.
Explanation
Correct Answers
B:The reason it is necessary is: Gradual deployment reduces blast radius only if meaningful indicators and rollback decisions are predefined. For this situation: A bank deploys a new classifier to all customers after testing only one branch and later discovers disparate denials.
D:The reason it is necessary is: Gradual deployment reduces blast radius only if meaningful indicators and rollback decisions are predefined. The corrective action is: Use risk-aware canary populations, guardrail metrics, stop criteria and tested rollback or kill switch.
Incorrect Answers
A:This attractive shortcut does not resolve the Canary releases and model rollback thresholds weakness: Call a deployment canary controlled when traffic fraction is small but no adverse-impact trigger is monitored. For this situation: A bank deploys a new classifier to all customers after testing only one branch and later discovers disparate denials.
C:This option concerns Model poisoning incident response and rollback, which is a related but different AI governance concern. It does not establish control effectiveness for Canary releases and model rollback thresholds.
E:This option concerns Model version change impact classification, which is a related but different AI governance concern. For this situation: A bank deploys a new classifier to all customers after testing only one branch and later discovers disparate denials.
