Isaca AAIA Practice Test 91
Question 900
An independent AI auditor reviews this case: A prompt logging pipeline accidentally sends employee personal identifiers to a public analytics workspace. Which THREE elements together provide a defensible AI audit conclusion: the corrective control, the evidence to test it, and a verified outcome? Select THREE answers.
- Review access traces, affected documents, tenant identity, logs, containment steps and privacy leadership decisions.
- Scope and notifications are accurate, exposure is blocked and controls are retested before reopening.
- Delete all logs immediately to hide exposed data and mark the incident resolved without determining scope.
- Contain runaway usage with quotas, timeout and backpressure, investigate access and validate recovery.
- Inspect impact inventory, rollback record, retesting, business communications and lessons-learned owners.
- Contain disclosure, preserve evidence safely, scope affected data and follow jurisdiction-specific notification assessment.
Explanation
Correct Answers
A:The reason it is necessary is: AI privacy incidents require ordinary breach governance plus model, retrieval and prompt-specific evidence. For this situation: A prompt logging pipeline accidentally sends employee personal identifiers to a public analytics workspace.
B:This is an observable validation result for AI privacy incident preservation and escalation, not just management assurance: Scope and notifications are accurate, exposure is blocked and controls are retested before reopening.
F:This response fits the risk associated with AI privacy incident preservation and escalation: Contain disclosure, preserve evidence safely, scope affected data and follow jurisdiction-specific notification assessment.
Incorrect Answers
C:It misses the key control principle: AI privacy incidents require ordinary breach governance plus model, retrieval and prompt-specific evidence.
D:This option concerns AI availability and unbounded consumption incident, which is a related but different AI governance concern. For this situation: A prompt logging pipeline accidentally sends employee personal identifiers to a public analytics workspace.
E:This option concerns AI model rollback and stakeholder communication, which is a related but different AI governance concern. A proportionate response is: Contain disclosure, preserve evidence safely, scope affected data and follow jurisdiction-specific notification assessment.
Question 901
A governance and assurance review uncovers the following: An external training feed abruptly changes outputs in a fraud engine and the release pipeline lacks prior checkpoints. What is the MOST appropriate risk-based audit recommendation that addresses the root control weakness? Select ONE best answer.
- Quarantine suspicious inputs and models, preserve provenance, revert to validated artifacts and investigate contamination.
- Disable or restrict affected tools, preserve transaction records and verify downstream changes before re-enabling.
- Delete suspicious records from live storage but leave poisoned model weights running as if unaffected.
- Coordinate technical recovery with affected stakeholder review, corrective communication and validated post-incident controls.
Explanation
Correct Answer
A:The reason it is necessary is: Poisoning response must address both malicious input and any model or derived features trained on it. The acceptable verified outcome is: Validated clean versions restore approved behavior and compromised data cannot silently retrain new releases.
Incorrect Answers
B:This option concerns AI agent unauthorized action incident, which is a related but different AI governance concern. It does not establish control effectiveness for Model poisoning incident response and rollback.
C:This attractive shortcut does not resolve the Model poisoning incident response and rollback weakness: Delete suspicious records from live storage but leave poisoned model weights running as if unaffected.
D:This option concerns AI model rollback and stakeholder communication, which is a related but different AI governance concern. It does not establish control effectiveness for Model poisoning incident response and rollback.
Question 902
A risk-based audit engagement examines this operating condition: An external training feed abruptly changes outputs in a fraud engine and the release pipeline lacks prior checkpoints. Which principle BEST explains why the observed practice is a material governance or control concern? Select ONE best answer.
- Poisoning response must address both malicious input and any model or derived features trained on it.
- Delete suspicious records from live storage but leave poisoned model weights running as if unaffected.
- Agentic incidents involve real external actions; merely correcting generated text may not undo the harm.
- An AI incident ends only after decision impacts, service restoration and recurrence risk are addressed.
Explanation
Correct Answer
A:The acceptable verified outcome is: Validated clean versions restore approved behavior and compromised data cannot silently retrain new releases. In this case: An external training feed abruptly changes outputs in a fraud engine and the release pipeline lacks prior checkpoints.
Incorrect Answers
B:This attractive shortcut does not resolve the Model poisoning incident response and rollback weakness: Delete suspicious records from live storage but leave poisoned model weights running as if unaffected. For this situation: An external training feed abruptly changes outputs in a fraud engine and the release pipeline lacks prior checkpoints.
C:This option concerns AI agent unauthorized action incident, which is a related but different AI governance concern. For this situation: An external training feed abruptly changes outputs in a fraud engine and the release pipeline lacks prior checkpoints.
D:This option concerns AI model rollback and stakeholder communication, which is a related but different AI governance concern. For this situation: An external training feed abruptly changes outputs in a fraud engine and the release pipeline lacks prior checkpoints.
Question 903
An audit committee asks for a defensible assessment of this situation: An external training feed abruptly changes outputs in a fraud engine and the release pipeline lacks prior checkpoints. What evidence should the auditor examine FIRST to substantiate or challenge the control owner’s claim? Select ONE best answer.
- Inspect request counts, context sizes, token or compute cost, caller attribution and saturation events.
- Inspect ingestion bursts, source identities, dataset hashes, registry links and clean restore evidence.
- Inspect canary assignment, segment KPIs, error thresholds, baseline model artifacts and rollback drill.
- Delete suspicious records from live storage but leave poisoned model weights running as if unaffected.
Explanation
Correct Answer
B:The reason it is necessary is: Poisoning response must address both malicious input and any model or derived features trained on it. In this case: An external training feed abruptly changes outputs in a fraud engine and the release pipeline lacks prior checkpoints.
Incorrect Answers
A:This option concerns AI availability and unbounded consumption incident, which is a related but different AI governance concern. It does not establish control effectiveness for Model poisoning incident response and rollback.
C:This option concerns Canary releases and model rollback thresholds, which is a related but different AI governance concern. It does not establish control effectiveness for Model poisoning incident response and rollback.
D:This attractive shortcut does not resolve the Model poisoning incident response and rollback weakness: Delete suspicious records from live storage but leave poisoned model weights running as if unaffected. The auditor should instead examine: Inspect ingestion bursts, source identities, dataset hashes, registry links and clean restore evidence.
Question 904
During an AI controls assessment, the team learns that an external training feed abruptly changes outputs in a fraud engine and the release pipeline lacks prior checkpoints. After management addresses the issue, which result MOST convincingly demonstrates effective remediation? Select ONE best answer.
- Only signed approved model-data combinations reach production and privilege tests fail closed.
- Delete suspicious records from live storage but leave poisoned model weights running as if unaffected.
- Validated clean versions restore approved behavior and compromised data cannot silently retrain new releases.
- Affected decisions are reviewed, relevant stakeholders notified and preventive controls verified.
Explanation
Correct Answer
C:The reason it is necessary is: Poisoning response must address both malicious input and any model or derived features trained on it. This is an observable validation result for Model poisoning incident response and rollback, not just management assurance: Validated clean versions restore approved behavior and compromised data cannot silently retrain new releases.
Incorrect Answers
A:This option concerns AI data and model CI/CD separation, which is a related but different AI governance concern. It does not establish control effectiveness for Model poisoning incident response and rollback.
B:This attractive shortcut does not resolve the Model poisoning incident response and rollback weakness: Delete suspicious records from live storage but leave poisoned model weights running as if unaffected. Effective closure requires: Validated clean versions restore approved behavior and compromised data cannot silently retrain new releases.
D:This option concerns AI model rollback and stakeholder communication, which is a related but different AI governance concern. The decisive principle here is: Poisoning response must address both malicious input and any model or derived features trained on it.
Question 905
An independent AI auditor reviews this case: A public feedback campaign skews a ranking model and customer outcomes deteriorate across regions. What is the MOST appropriate risk-based audit recommendation that addresses the root control weakness? Select ONE best answer.
- Separate development, test and production roles with controlled promotion, scoped credentials and audit trails.
- Delete suspicious records from live storage but leave poisoned model weights running as if unaffected.
- Quarantine suspicious inputs and models, preserve provenance, revert to validated artifacts and investigate contamination.
- Use risk-aware canary populations, guardrail metrics, stop criteria and tested rollback or kill switch.
Explanation
Correct Answer
C:The reason it is necessary is: Poisoning response must address both malicious input and any model or derived features trained on it. This is the proportionate control response to Model poisoning incident response and rollback: Quarantine suspicious inputs and models, preserve provenance, revert to validated artifacts and investigate contamination.
Incorrect Answers
A:This option concerns AI data and model CI/CD separation, which is a related but different AI governance concern. The decisive principle here is: Poisoning response must address both malicious input and any model or derived features trained on it.
B:This attractive shortcut does not resolve the Model poisoning incident response and rollback weakness: Delete suspicious records from live storage but leave poisoned model weights running as if unaffected. In this case: A public feedback campaign skews a ranking model and customer outcomes deteriorate across regions.
D:This option concerns Canary releases and model rollback thresholds, which is a related but different AI governance concern. The decisive principle here is: Poisoning response must address both malicious input and any model or derived features trained on it.
Question 906
A governance and assurance review uncovers the following: A public feedback campaign skews a ranking model and customer outcomes deteriorate across regions. Which principle BEST explains why the observed practice is a material governance or control concern? Select ONE best answer.
- Poisoning response must address both malicious input and any model or derived features trained on it.
- AI delivery involves models, data and code, so conventional app deployment approval alone may not cover behavior changes.
- AI privacy incidents require ordinary breach governance plus model, retrieval and prompt-specific evidence.
- Delete suspicious records from live storage but leave poisoned model weights running as if unaffected.
Explanation
Correct Answer
A:The acceptable verified outcome is: Validated clean versions restore approved behavior and compromised data cannot silently retrain new releases. Independent audit evidence includes: Inspect ingestion bursts, source identities, dataset hashes, registry links and clean restore evidence.
Incorrect Answers
B:This option concerns AI data and model CI/CD separation, which is a related but different AI governance concern. For this situation: A public feedback campaign skews a ranking model and customer outcomes deteriorate across regions.
C:This option concerns AI privacy incident preservation and escalation, which is a related but different AI governance concern. It does not establish control effectiveness for Model poisoning incident response and rollback.
D:This attractive shortcut does not resolve the Model poisoning incident response and rollback weakness: Delete suspicious records from live storage but leave poisoned model weights running as if unaffected. For this situation: A public feedback campaign skews a ranking model and customer outcomes deteriorate across regions.
Question 907
A risk-based audit engagement examines this operating condition: A public feedback campaign skews a ranking model and customer outcomes deteriorate across regions. What evidence should the auditor examine FIRST to substantiate or challenge the control owner’s claim? Select ONE best answer.
- Inspect impact inventory, rollback record, retesting, business communications and lessons-learned owners.
- Delete suspicious records from live storage but leave poisoned model weights running as if unaffected.
- Review access traces, affected documents, tenant identity, logs, containment steps and privacy leadership decisions.
- Inspect ingestion bursts, source identities, dataset hashes, registry links and clean restore evidence.
Explanation
Correct Answer
D:The reason it is necessary is: Poisoning response must address both malicious input and any model or derived features trained on it. This evidence directly tests the asserted control for Model poisoning incident response and rollback: Inspect ingestion bursts, source identities, dataset hashes, registry links and clean restore evidence.
Incorrect Answers
A:This option concerns AI model rollback and stakeholder communication, which is a related but different AI governance concern. The audit must corroborate: Inspect ingestion bursts, source identities, dataset hashes, registry links and clean restore evidence.
B:This attractive shortcut does not resolve the Model poisoning incident response and rollback weakness: Delete suspicious records from live storage but leave poisoned model weights running as if unaffected. Quarantine suspicious inputs and models, preserve provenance, revert to validated artifacts and investigate contamination.
C:This option concerns AI privacy incident preservation and escalation, which is a related but different AI governance concern. The decisive principle here is: Poisoning response must address both malicious input and any model or derived features trained on it.
Question 908
An audit committee asks for a defensible assessment of this situation: A public feedback campaign skews a ranking model and customer outcomes deteriorate across regions. After management addresses the issue, which result MOST convincingly demonstrates effective remediation? Select ONE best answer.
- Delete suspicious records from live storage but leave poisoned model weights running as if unaffected.
- Validated clean versions restore approved behavior and compromised data cannot silently retrain new releases.
- Affected decisions are reviewed, relevant stakeholders notified and preventive controls verified.
- An independent rerun can identify the exact source data and material transformation decisions for a sampled model version.
Explanation
Correct Answer
B:The reason it is necessary is: Poisoning response must address both malicious input and any model or derived features trained on it. Independent audit evidence includes: Inspect ingestion bursts, source identities, dataset hashes, registry links and clean restore evidence.
Incorrect Answers
A:It misses the key control principle: Poisoning response must address both malicious input and any model or derived features trained on it.
C:This option concerns AI model rollback and stakeholder communication, which is a related but different AI governance concern. For this situation: A public feedback campaign skews a ranking model and customer outcomes deteriorate across regions.
D: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 Model poisoning incident response and rollback.
Question 909
During an AI controls assessment, the team learns that a security classifier's trusted training labels are replaced with malicious values just before nightly retraining. What is the MOST appropriate risk-based audit recommendation that addresses the root control weakness? Select ONE best answer.
- Quarantine suspicious inputs and models, preserve provenance, revert to validated artifacts and investigate contamination.
- Contain runaway usage with quotas, timeout and backpressure, investigate access and validate recovery.
- Activate an AI-specific incident plan to constrain outputs, preserve evidence, notify responsible owners and restore validated service.
- Delete suspicious records from live storage but leave poisoned model weights running as if unaffected.
Explanation
Correct Answer
A:The reason it is necessary is: Poisoning response must address both malicious input and any model or derived features trained on it. In this case: A security classifier's trusted training labels are replaced with malicious values just before nightly retraining.
Incorrect Answers
B:This option concerns AI availability and unbounded consumption incident, which is a related but different AI governance concern. The decisive principle here is: Poisoning response must address both malicious input and any model or derived features trained on it.
C:This option concerns Response to harmful AI-generated advice, which is a related but different AI governance concern. It does not establish control effectiveness for Model poisoning incident response and rollback.
D:This attractive shortcut does not resolve the Model poisoning incident response and rollback weakness: Delete suspicious records from live storage but leave poisoned model weights running as if unaffected. In this case: A security classifier's trusted training labels are replaced with malicious values just before nightly retraining.
Question 910
An independent AI auditor reviews this case: A security classifier's trusted training labels are replaced with malicious values just before nightly retraining. Which TWO complementary elements should the auditor seek: an effective corrective control and the most probative independent evidence? Select TWO answers.
- Quarantine suspicious inputs and models, preserve provenance, revert to validated artifacts and investigate contamination.
- Inspect ingestion bursts, source identities, dataset hashes, registry links and clean restore evidence.
- Delete suspicious records from live storage but leave poisoned model weights running as if unaffected.
- Inspect tool invocation audit, actor scopes, approval checks, external transaction state and rollback capability.
- Separate development, test and production roles with controlled promotion, scoped credentials and audit trails.
Explanation
Correct Answers
A:The reason it is necessary is: Poisoning response must address both malicious input and any model or derived features trained on it. This response fits the risk associated with Model poisoning incident response and rollback: Quarantine suspicious inputs and models, preserve provenance, revert to validated artifacts and investigate contamination.
B:The reason it is necessary is: Poisoning response must address both malicious input and any model or derived features trained on it. The corrective action is: Quarantine suspicious inputs and models, preserve provenance, revert to validated artifacts and investigate contamination.
Incorrect Answers
C:This attractive shortcut does not resolve the Model poisoning incident response and rollback weakness: Delete suspicious records from live storage but leave poisoned model weights running as if unaffected. For this situation: A security classifier's trusted training labels are replaced with malicious values just before nightly retraining.
D:This option concerns AI agent unauthorized action incident, which is a related but different AI governance concern. The decisive principle here is: Poisoning response must address both malicious input and any model or derived features trained on it.
E:This option concerns AI data and model CI/CD separation, which is a related but different AI governance concern. The audit must corroborate: Inspect ingestion bursts, source identities, dataset hashes, registry links and clean restore evidence.
