ISACA Advanced in AI Audit AAIA Practice Test 8
Question 70
A chatbot threat register lists only malware while prompt injection can disclose retrieved documents. From an AI audit perspective, which corrective action is the MOST direct response to the weakness described, rather than a useful control for some different AI risk?
- Map retrieval, prompts, tools, outputs and trust boundaries into the threat assessment
- Classify each deployment use case separately based on actual harm pathways
- Map external model and data dependencies to failure modes, recovery plans and substitutes
- Enforce risk-acceptance authority levels, expiry and independent challenge for high impact
Explanation
Correct Answer
A:Correct. Map retrieval, prompts, tools, outputs and trust boundaries into the threat assessment is aimed at the actual failure: ai-specific attack surfaces were excluded from threat modeling. The audit should subsequently obtain threat model and tested prompt-injection paths with tracked mitigations to test execution, not merely the existence of a policy.
Incorrect Answers
B:“Classify each deployment use case separately based on actual harm pathways” is a control for risk classification ignores decision materiality and affected-person rights, but the audit finding here is aI-specific attack surfaces were excluded from threat modeling. The relevant measure is “Map retrieval, prompts, tools, outputs and trust boundaries into the threat assessment”.
C:“Map external model and data dependencies to failure modes, recovery plans and substitutes” is a control for aI supply-chain concentration risk is missing from business impact assessment, but the audit finding here is aI-specific attack surfaces were excluded from threat modeling. The relevant measure is “Map retrieval, prompts, tools, outputs and trust boundaries into the threat assessment”.
D:“Enforce risk-acceptance authority levels, expiry and independent challenge for high impact” is a control for residual risk acceptance exceeds defined decision rights, but the audit finding here is aI-specific attack surfaces were excluded from threat modeling. The relevant measure is “Map retrieval, prompts, tools, outputs and trust boundaries into the threat assessment”.
Question 71
In an audit of AI Risk Management, managers say this measure is live: Map retrieval, prompts, tools, outputs and trust boundaries into the threat assessment. They claim it corrected a material audit gap. What direct operating evidence would allow the auditor to substantiate that particular claim across a defensible sample?
- Use-case risk register with distinct controls and documented tier approvals
- Dependency map and tested fallback to an approved alternative provider
- Approval trail showing delegated signatories and expired exceptions revisited
- Threat model and tested prompt-injection paths with tracked mitigations
Explanation
Correct Answer
D:Correct. The proposed records—threat model and tested prompt-injection paths with tracked mitigations—are directly tied to the asserted safeguard and allow its actual operation to be corroborated. The gap being tested is ai-specific attack surfaces were excluded from threat modeling.
Incorrect Answers
A:“Use-case risk register with distinct controls and documented tier approvals” would test risk classification ignores decision materiality and affected-person rights, not the present safeguard against aI-specific attack surfaces were excluded from threat modeling. The expected audit evidence is “Threat model and tested prompt-injection paths with tracked mitigations”.
B:“Dependency map and tested fallback to an approved alternative provider” would test aI supply-chain concentration risk is missing from business impact assessment, not the present safeguard against aI-specific attack surfaces were excluded from threat modeling. The expected audit evidence is “Threat model and tested prompt-injection paths with tracked mitigations”.
C:“Approval trail showing delegated signatories and expired exceptions revisited” would test residual risk acceptance exceeds defined decision rights, not the present safeguard against aI-specific attack surfaces were excluded from threat modeling. The expected audit evidence is “Threat model and tested prompt-injection paths with tracked mitigations”.
Question 72
In a review of AI Risk Management, investigators cannot establish the expected control proof: threat model and tested prompt-injection paths with tracked mitigations. The underlying event was that a chatbot threat register lists only malware while prompt injection can disclose retrieved documents. Which control weakness is the MOST likely cause requiring focused audit work?
- Risk classification ignores decision materiality and affected-person rights
- AI supply-chain concentration risk is missing from business impact assessment
- AI-specific attack surfaces were excluded from threat modeling
- Residual risk acceptance exceeds defined decision rights
Explanation
Correct Answer
C:Correct. The observed condition specifically indicates ai-specific attack surfaces were excluded from threat modeling. A meaningful audit response is to verify this diagnosis and then recommend map retrieval, prompts, tools, outputs and trust boundaries into the threat assessment.
Incorrect Answers
A:“Risk classification ignores decision materiality and affected-person rights” concerns a separate audit weakness, not the missing assurance described by aI-specific attack surfaces were excluded from threat modeling. The correct cause is “AI-specific attack surfaces were excluded from threat modeling”.
B:“AI supply-chain concentration risk is missing from business impact assessment” concerns a separate audit weakness, not the missing assurance described by aI-specific attack surfaces were excluded from threat modeling. The correct cause is “AI-specific attack surfaces were excluded from threat modeling”.
D:“Residual risk acceptance exceeds defined decision rights” concerns a separate audit weakness, not the missing assurance described by aI-specific attack surfaces were excluded from threat modeling. The correct cause is “AI-specific attack surfaces were excluded from threat modeling”.
Question 73
A management report marks all AI risk green because controls exist but none were tested. From an AI audit perspective, which corrective action is the MOST direct response to the weakness described, rather than a useful control for some different AI risk?
- Set quantitative appetite triggers, cumulative exposure limits and escalation owners
- Report inherent risk, control test results and residual risk with evidence and uncertainty
- Threat-model tool permissions and require least privilege plus outbound data restrictions
- Score risk against impact, autonomy, vulnerability and exposure rather than data value alone
Explanation
Correct Answer
B:Correct. Report inherent risk, control test results and residual risk with evidence and uncertainty is aimed at the actual failure: control design was mistaken for demonstrated operating effectiveness. The audit should subsequently obtain risk dashboard linked to effectiveness samples and residual-risk approvals to test execution, not merely the existence of a policy.
Incorrect Answers
A:“Set quantitative appetite triggers, cumulative exposure limits and escalation owners” is a control for exceptions and aggregation limits are absent from AI risk appetite implementation, but the audit finding here is control design was mistaken for demonstrated operating effectiveness. The relevant measure is “Report inherent risk, control test results and residual risk with evidence and uncertainty”.
C:“Threat-model tool permissions and require least privilege plus outbound data restrictions” is a control for tool-call exfiltration risk was omitted from AI misuse scenarios, but the audit finding here is control design was mistaken for demonstrated operating effectiveness. The relevant measure is “Report inherent risk, control test results and residual risk with evidence and uncertainty”.
D:“Score risk against impact, autonomy, vulnerability and exposure rather than data value alone” is a control for inherent risk scoring excludes safety impact and severity of affected decisions, but the audit finding here is control design was mistaken for demonstrated operating effectiveness. The relevant measure is “Report inherent risk, control test results and residual risk with evidence and uncertainty”.
Question 74
In an audit of AI Risk Management, managers say this measure is live: Report inherent risk, control test results and residual risk with evidence and uncertainty. They claim it corrected a material audit gap. What direct operating evidence would allow the auditor to substantiate that particular claim across a defensible sample?
- Risk dashboard linked to effectiveness samples and residual-risk approvals
- Breach logs, consolidated exposures and minutes of timely risk committee action
- Permission matrix, red-team exfiltration test and blocked outbound traces
- Scoring rationale, safety severity scenarios and independent challenge records
Explanation
Correct Answer
A:Correct. The proposed records—risk dashboard linked to effectiveness samples and residual-risk approvals—are directly tied to the asserted safeguard and allow its actual operation to be corroborated. The gap being tested is control design was mistaken for demonstrated operating effectiveness.
Incorrect Answers
B:“Breach logs, consolidated exposures and minutes of timely risk committee action” would test exceptions and aggregation limits are absent from AI risk appetite implementation, not the present safeguard against control design was mistaken for demonstrated operating effectiveness. The expected audit evidence is “Risk dashboard linked to effectiveness samples and residual-risk approvals”.
C:“Permission matrix, red-team exfiltration test and blocked outbound traces” would test tool-call exfiltration risk was omitted from AI misuse scenarios, not the present safeguard against control design was mistaken for demonstrated operating effectiveness. The expected audit evidence is “Risk dashboard linked to effectiveness samples and residual-risk approvals”.
D:“Scoring rationale, safety severity scenarios and independent challenge records” would test inherent risk scoring excludes safety impact and severity of affected decisions, not the present safeguard against control design was mistaken for demonstrated operating effectiveness. The expected audit evidence is “Risk dashboard linked to effectiveness samples and residual-risk approvals”.
Question 75
In a review of AI Risk Management, investigators cannot establish the expected control proof: risk dashboard linked to effectiveness samples and residual-risk approvals. The underlying event was that a management report marks all AI risk green because controls exist but none were tested. Which control weakness is the MOST likely cause requiring focused audit work?
- Exceptions and aggregation limits are absent from AI risk appetite implementation
- Tool-call exfiltration risk was omitted from AI misuse scenarios
- Inherent risk scoring excludes safety impact and severity of affected decisions
- Control design was mistaken for demonstrated operating effectiveness
Explanation
Correct Answer
D:Correct. The observed condition specifically indicates control design was mistaken for demonstrated operating effectiveness. A meaningful audit response is to verify this diagnosis and then recommend report inherent risk, control test results and residual risk with evidence and uncertainty.
Incorrect Answers
A:“Exceptions and aggregation limits are absent from AI risk appetite implementation” concerns a separate audit weakness, not the missing assurance described by control design was mistaken for demonstrated operating effectiveness. The correct cause is “Control design was mistaken for demonstrated operating effectiveness”.
B:“Tool-call exfiltration risk was omitted from AI misuse scenarios” concerns a separate audit weakness, not the missing assurance described by control design was mistaken for demonstrated operating effectiveness. The correct cause is “Control design was mistaken for demonstrated operating effectiveness”.
C:“Inherent risk scoring excludes safety impact and severity of affected decisions” concerns a separate audit weakness, not the missing assurance described by control design was mistaken for demonstrated operating effectiveness. The correct cause is “Control design was mistaken for demonstrated operating effectiveness”.
Question 76
One AI tool is used in both recruiting and cafeteria menu planning but assigned the same risk tier. From an AI audit perspective, which corrective action is the MOST direct response to the weakness described, rather than a useful control for some different AI risk?
- Establish disaggregated performance/fairness KRIs and periodic reassessment triggers
- Re-rate affected risk scenarios and validation evidence after material base-model changes
- Classify each deployment use case separately based on actual harm pathways
- Map retrieval, prompts, tools, outputs and trust boundaries into the threat assessment
Explanation
Correct Answer
C:Correct. Classify each deployment use case separately based on actual harm pathways is aimed at the actual failure: risk classification ignores decision materiality and affected-person rights. The audit should subsequently obtain use-case risk register with distinct controls and documented tier approvals to test execution, not merely the existence of a policy.
Incorrect Answers
A:“Establish disaggregated performance/fairness KRIs and periodic reassessment triggers” is a control for fairness risk is being treated as static despite population drift, but the audit finding here is risk classification ignores decision materiality and affected-person rights. The relevant measure is “Classify each deployment use case separately based on actual harm pathways”.
B:“Re-rate affected risk scenarios and validation evidence after material base-model changes” is a control for change events do not trigger risk reassessment, but the audit finding here is risk classification ignores decision materiality and affected-person rights. The relevant measure is “Classify each deployment use case separately based on actual harm pathways”.
D:“Map retrieval, prompts, tools, outputs and trust boundaries into the threat assessment” is a control for aI-specific attack surfaces were excluded from threat modeling, but the audit finding here is risk classification ignores decision materiality and affected-person rights. The relevant measure is “Classify each deployment use case separately based on actual harm pathways”.
Question 77
In an audit of AI Risk Management, managers say this measure is live: Classify each deployment use case separately based on actual harm pathways. They claim it corrected a material audit gap. What direct operating evidence would allow the auditor to substantiate that particular claim across a defensible sample?
- Quarterly cohort metrics, drift findings and closed remediation tickets
- Use-case risk register with distinct controls and documented tier approvals
- Version-linked risk register updates, retest results and formal approvals
- Threat model and tested prompt-injection paths with tracked mitigations
Explanation
Correct Answer
B:Correct. The proposed records—use-case risk register with distinct controls and documented tier approvals—are directly tied to the asserted safeguard and allow its actual operation to be corroborated. The gap being tested is risk classification ignores decision materiality and affected-person rights.
Incorrect Answers
A:“Quarterly cohort metrics, drift findings and closed remediation tickets” would test fairness risk is being treated as static despite population drift, not the present safeguard against risk classification ignores decision materiality and affected-person rights. The expected audit evidence is “Use-case risk register with distinct controls and documented tier approvals”.
C:“Version-linked risk register updates, retest results and formal approvals” would test change events do not trigger risk reassessment, not the present safeguard against risk classification ignores decision materiality and affected-person rights. The expected audit evidence is “Use-case risk register with distinct controls and documented tier approvals”.
D:“Threat model and tested prompt-injection paths with tracked mitigations” would test aI-specific attack surfaces were excluded from threat modeling, not the present safeguard against risk classification ignores decision materiality and affected-person rights. The expected audit evidence is “Use-case risk register with distinct controls and documented tier approvals”.
Question 78
In a review of AI Risk Management, investigators cannot establish the expected control proof: use-case risk register with distinct controls and documented tier approvals. The underlying event was that one AI tool is used in both recruiting and cafeteria menu planning but assigned the same risk tier. Which control weakness is the MOST likely cause requiring focused audit work?
- Risk classification ignores decision materiality and affected-person rights
- Fairness risk is being treated as static despite population drift
- Change events do not trigger risk reassessment
- AI-specific attack surfaces were excluded from threat modeling
Explanation
Correct Answer
A:Correct. The observed condition specifically indicates risk classification ignores decision materiality and affected-person rights. A meaningful audit response is to verify this diagnosis and then recommend classify each deployment use case separately based on actual harm pathways.
Incorrect Answers
B:“Fairness risk is being treated as static despite population drift” concerns a separate audit weakness, not the missing assurance described by risk classification ignores decision materiality and affected-person rights. The correct cause is “Risk classification ignores decision materiality and affected-person rights”.
C:“Change events do not trigger risk reassessment” concerns a separate audit weakness, not the missing assurance described by risk classification ignores decision materiality and affected-person rights. The correct cause is “Risk classification ignores decision materiality and affected-person rights”.
D:“AI-specific attack surfaces were excluded from threat modeling” concerns a separate audit weakness, not the missing assurance described by risk classification ignores decision materiality and affected-person rights. The correct cause is “Risk classification ignores decision materiality and affected-person rights”.
Question 79
A subsidiary repeatedly overrides an error threshold without escalating to group risk leadership. From an AI audit perspective, which corrective action is the MOST direct response to the weakness described, rather than a useful control for some different AI risk?
- Map external model and data dependencies to failure modes, recovery plans and substitutes
- Include foreseeable misuse, malicious insiders and automation bias in scenario workshops
- Report inherent risk, control test results and residual risk with evidence and uncertainty
- Set quantitative appetite triggers, cumulative exposure limits and escalation owners
Explanation
Correct Answer
D:Correct. Set quantitative appetite triggers, cumulative exposure limits and escalation owners is aimed at the actual failure: exceptions and aggregation limits are absent from ai risk appetite implementation. The audit should subsequently obtain breach logs, consolidated exposures and minutes of timely risk committee action to test execution, not merely the existence of a policy.
Incorrect Answers
A:“Map external model and data dependencies to failure modes, recovery plans and substitutes” is a control for aI supply-chain concentration risk is missing from business impact assessment, but the audit finding here is exceptions and aggregation limits are absent from AI risk appetite implementation. The relevant measure is “Set quantitative appetite triggers, cumulative exposure limits and escalation owners”.
B:“Include foreseeable misuse, malicious insiders and automation bias in scenario workshops” is a control for risk discovery is limited to expected user behavior, but the audit finding here is exceptions and aggregation limits are absent from AI risk appetite implementation. The relevant measure is “Set quantitative appetite triggers, cumulative exposure limits and escalation owners”.
C:“Report inherent risk, control test results and residual risk with evidence and uncertainty” is a control for control design was mistaken for demonstrated operating effectiveness, but the audit finding here is exceptions and aggregation limits are absent from AI risk appetite implementation. The relevant measure is “Set quantitative appetite triggers, cumulative exposure limits and escalation owners”.
Question 80
In an audit of AI Risk Management, managers say this measure is live: Set quantitative appetite triggers, cumulative exposure limits and escalation owners. They claim it corrected a material audit gap. What direct operating evidence would allow the auditor to substantiate that particular claim across a defensible sample?
- Dependency map and tested fallback to an approved alternative provider
- Recorded misuse scenarios, red-team cases and completed owner treatment decisions
- Breach logs, consolidated exposures and minutes of timely risk committee action
- Risk dashboard linked to effectiveness samples and residual-risk approvals
Explanation
Correct Answer
C:Correct. The proposed records—breach logs, consolidated exposures and minutes of timely risk committee action—are directly tied to the asserted safeguard and allow its actual operation to be corroborated. The gap being tested is exceptions and aggregation limits are absent from ai risk appetite implementation.
Incorrect Answers
A:“Dependency map and tested fallback to an approved alternative provider” would test aI supply-chain concentration risk is missing from business impact assessment, not the present safeguard against exceptions and aggregation limits are absent from AI risk appetite implementation. The expected audit evidence is “Breach logs, consolidated exposures and minutes of timely risk committee action”.
B:“Recorded misuse scenarios, red-team cases and completed owner treatment decisions” would test risk discovery is limited to expected user behavior, not the present safeguard against exceptions and aggregation limits are absent from AI risk appetite implementation. The expected audit evidence is “Breach logs, consolidated exposures and minutes of timely risk committee action”.
D:“Risk dashboard linked to effectiveness samples and residual-risk approvals” would test control design was mistaken for demonstrated operating effectiveness, not the present safeguard against exceptions and aggregation limits are absent from AI risk appetite implementation. The expected audit evidence is “Breach logs, consolidated exposures and minutes of timely risk committee action”.