ISACA Advanced in AI Audit AAIA Practice Test 37
Question 360
In a review of Threats and Vulnerabilities Specific to AI, investigators cannot establish the expected control proof: xss test payload blocked by renderer and browser policy. The underlying event was that a chatbot output is rendered as HTML by an internal dashboard and injected script executes in reviewers' browsers. Which control weakness is the MOST likely cause requiring focused audit work?
- Adaptive jailbreak testing and abuse monitoring were inadequate
- Feedback data lacks integrity and poisoning detection
- Model output is being trusted as executable presentation code
- Membership-inference exposure was not included in privacy threat assessment
Explanation
Correct Answer
C:Correct. The observed condition specifically indicates model output is being trusted as executable presentation code. A meaningful audit response is to verify this diagnosis and then recommend encode untrusted model output and apply strict rendering/content security controls.
Incorrect Answers
A:“Adaptive jailbreak testing and abuse monitoring were inadequate” concerns a separate audit weakness, not the missing assurance described by model output is being trusted as executable presentation code. The correct cause is “Model output is being trusted as executable presentation code”.
B:“Feedback data lacks integrity and poisoning detection” concerns a separate audit weakness, not the missing assurance described by model output is being trusted as executable presentation code. The correct cause is “Model output is being trusted as executable presentation code”.
D:“Membership-inference exposure was not included in privacy threat assessment” concerns a separate audit weakness, not the missing assurance described by model output is being trusted as executable presentation code. The correct cause is “Model output is being trusted as executable presentation code”.
Question 361
An AI retrieval service shares one vector index across unrelated tenants without filter enforcement. 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?
- Isolate untrusted content and enforce tool-level authorization and egress checks
- Enforce server-side tenant authorization on every vector search
- Sign and verify model weight hashes at promotion and load
- Verify provenance and execute untrusted packages in restricted build environments
Explanation
Correct Answer
B:Correct. Enforce server-side tenant authorization on every vector search is aimed at the actual failure: tenant isolation is enforced only by client-provided retrieval hints. The audit should subsequently obtain cross-tenant negative queries and datastore policy test results to test execution, not merely the existence of a policy.
Incorrect Answers
A:“Isolate untrusted content and enforce tool-level authorization and egress checks” is a control for retrieved content is allowed to cross the instruction trust boundary, but the audit finding here is tenant isolation is enforced only by client-provided retrieval hints. The relevant measure is “Enforce server-side tenant authorization on every vector search”.
C:“Sign and verify model weight hashes at promotion and load” is a control for model artifacts themselves lack integrity controls independent of container signing, but the audit finding here is tenant isolation is enforced only by client-provided retrieval hints. The relevant measure is “Enforce server-side tenant authorization on every vector search”.
D:“Verify provenance and execute untrusted packages in restricted build environments” is a control for model supply-chain files were not scanned or sandboxed before execution, but the audit finding here is tenant isolation is enforced only by client-provided retrieval hints. The relevant measure is “Enforce server-side tenant authorization on every vector search”.
Question 362
In an audit of Threats and Vulnerabilities Specific to AI, managers say this measure is live: Enforce server-side tenant authorization on every vector search. 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?
- Cross-tenant negative queries and datastore policy test results
- Red-team poisoned-document results and blocked file-export calls
- Failed hash validation on tampered model artifact in a release test
- SBOM review, sandbox network trace and blocked installer execution
Explanation
Correct Answer
A:Correct. The proposed records—cross-tenant negative queries and datastore policy test results—are directly tied to the asserted safeguard and allow its actual operation to be corroborated. The gap being tested is tenant isolation is enforced only by client-provided retrieval hints.
Incorrect Answers
B:“Red-team poisoned-document results and blocked file-export calls” would test retrieved content is allowed to cross the instruction trust boundary, not the present safeguard against tenant isolation is enforced only by client-provided retrieval hints. The expected audit evidence is “Cross-tenant negative queries and datastore policy test results”.
C:“Failed hash validation on tampered model artifact in a release test” would test model artifacts themselves lack integrity controls independent of container signing, not the present safeguard against tenant isolation is enforced only by client-provided retrieval hints. The expected audit evidence is “Cross-tenant negative queries and datastore policy test results”.
D:“SBOM review, sandbox network trace and blocked installer execution” would test model supply-chain files were not scanned or sandboxed before execution, not the present safeguard against tenant isolation is enforced only by client-provided retrieval hints. The expected audit evidence is “Cross-tenant negative queries and datastore policy test results”.
Question 363
In a review of Threats and Vulnerabilities Specific to AI, investigators cannot establish the expected control proof: cross-tenant negative queries and datastore policy test results. The underlying event was that an AI retrieval service shares one vector index across unrelated tenants without filter enforcement. Which control weakness is the MOST likely cause requiring focused audit work?
- Retrieved content is allowed to cross the instruction trust boundary
- Model artifacts themselves lack integrity controls independent of container signing
- Model supply-chain files were not scanned or sandboxed before execution
- Tenant isolation is enforced only by client-provided retrieval hints
Explanation
Correct Answer
D:Correct. The observed condition specifically indicates tenant isolation is enforced only by client-provided retrieval hints. A meaningful audit response is to verify this diagnosis and then recommend enforce server-side tenant authorization on every vector search.
Incorrect Answers
A:“Retrieved content is allowed to cross the instruction trust boundary” concerns a separate audit weakness, not the missing assurance described by tenant isolation is enforced only by client-provided retrieval hints. The correct cause is “Tenant isolation is enforced only by client-provided retrieval hints”.
B:“Model artifacts themselves lack integrity controls independent of container signing” concerns a separate audit weakness, not the missing assurance described by tenant isolation is enforced only by client-provided retrieval hints. The correct cause is “Tenant isolation is enforced only by client-provided retrieval hints”.
C:“Model supply-chain files were not scanned or sandboxed before execution” concerns a separate audit weakness, not the missing assurance described by tenant isolation is enforced only by client-provided retrieval hints. The correct cause is “Tenant isolation is enforced only by client-provided retrieval hints”.
Question 364
An attacker systematically asks a content filter near-boundary prompts until the agent reveals protected material. 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?
- Apply rate limits, anomaly detection and output disclosure controls
- Restrict external tool destinations and inspect sensitive content before egress
- Add adversarial interaction monitoring, guardrail hardening and escalation
- Encode untrusted model output and apply strict rendering/content security controls
Explanation
Correct Answer
C:Correct. Add adversarial interaction monitoring, guardrail hardening and escalation is aimed at the actual failure: adaptive jailbreak testing and abuse monitoring were inadequate. The audit should subsequently obtain red-team adaptive prompt series and risk alert response evidence to test execution, not merely the existence of a policy.
Incorrect Answers
A:“Apply rate limits, anomaly detection and output disclosure controls” is a control for query-rate and extraction defenses are inadequate for model privacy, but the audit finding here is adaptive jailbreak testing and abuse monitoring were inadequate. The relevant measure is “Add adversarial interaction monitoring, guardrail hardening and escalation”.
B:“Restrict external tool destinations and inspect sensitive content before egress” is a control for tool egress and data-classification checks are missing, but the audit finding here is adaptive jailbreak testing and abuse monitoring were inadequate. The relevant measure is “Add adversarial interaction monitoring, guardrail hardening and escalation”.
D:“Encode untrusted model output and apply strict rendering/content security controls” is a control for model output is being trusted as executable presentation code, but the audit finding here is adaptive jailbreak testing and abuse monitoring were inadequate. The relevant measure is “Add adversarial interaction monitoring, guardrail hardening and escalation”.
Question 365
In an audit of Threats and Vulnerabilities Specific to AI, managers say this measure is live: Add adversarial interaction monitoring, guardrail hardening and escalation. 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?
- Extractive attack simulation, query telemetry and throttling events
- Red-team adaptive prompt series and risk alert response evidence
- Proxy block logs and tested denied tool requests containing secret strings
- XSS test payload blocked by renderer and browser policy
Explanation
Correct Answer
B:Correct. The proposed records—red-team adaptive prompt series and risk alert response evidence—are directly tied to the asserted safeguard and allow its actual operation to be corroborated. The gap being tested is adaptive jailbreak testing and abuse monitoring were inadequate.
Incorrect Answers
A:“Extractive attack simulation, query telemetry and throttling events” would test query-rate and extraction defenses are inadequate for model privacy, not the present safeguard against adaptive jailbreak testing and abuse monitoring were inadequate. The expected audit evidence is “Red-team adaptive prompt series and risk alert response evidence”.
C:“Proxy block logs and tested denied tool requests containing secret strings” would test tool egress and data-classification checks are missing, not the present safeguard against adaptive jailbreak testing and abuse monitoring were inadequate. The expected audit evidence is “Red-team adaptive prompt series and risk alert response evidence”.
D:“XSS test payload blocked by renderer and browser policy” would test model output is being trusted as executable presentation code, not the present safeguard against adaptive jailbreak testing and abuse monitoring were inadequate. The expected audit evidence is “Red-team adaptive prompt series and risk alert response evidence”.
Question 366
In a review of Threats and Vulnerabilities Specific to AI, investigators cannot establish the expected control proof: red-team adaptive prompt series and risk alert response evidence. The underlying event was that an attacker systematically asks a content filter near-boundary prompts until the agent reveals protected material. Which control weakness is the MOST likely cause requiring focused audit work?
- Adaptive jailbreak testing and abuse monitoring were inadequate
- Query-rate and extraction defenses are inadequate for model privacy
- Tool egress and data-classification checks are missing
- Model output is being trusted as executable presentation code
Explanation
Correct Answer
A:Correct. The observed condition specifically indicates adaptive jailbreak testing and abuse monitoring were inadequate. A meaningful audit response is to verify this diagnosis and then recommend add adversarial interaction monitoring, guardrail hardening and escalation.
Incorrect Answers
B:“Query-rate and extraction defenses are inadequate for model privacy” concerns a separate audit weakness, not the missing assurance described by adaptive jailbreak testing and abuse monitoring were inadequate. The correct cause is “Adaptive jailbreak testing and abuse monitoring were inadequate”.
C:“Tool egress and data-classification checks are missing” concerns a separate audit weakness, not the missing assurance described by adaptive jailbreak testing and abuse monitoring were inadequate. The correct cause is “Adaptive jailbreak testing and abuse monitoring were inadequate”.
D:“Model output is being trusted as executable presentation code” concerns a separate audit weakness, not the missing assurance described by adaptive jailbreak testing and abuse monitoring were inadequate. The correct cause is “Adaptive jailbreak testing and abuse monitoring were inadequate”.
Question 367
A fraud model suddenly rejects nearly every payment after a vendor feature outage. 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?
- Identify all impacted builds from lineage and quarantine affected artifacts
- Isolate generated deliverables, notify counsel and correct affected outputs with audit trail
- Preserve sensor, model, action and operator traces while engaging safety response
- Trigger circuit breaker, approved fallback and incident classification based on customer harm
Explanation
Correct Answer
D:Correct. Trigger circuit breaker, approved fallback and incident classification based on customer harm is aimed at the actual failure: ai incident runbooks do not define fallback and business-continuity response. The audit should subsequently obtain time-stamped failover exercise and incident response timeline to test execution, not merely the existence of a policy.
Incorrect Answers
A:“Identify all impacted builds from lineage and quarantine affected artifacts” is a control for model incidents do not trace data lineage and affected derivatives, but the audit finding here is aI incident runbooks do not define fallback and business-continuity response. The relevant measure is “Trigger circuit breaker, approved fallback and incident classification based on customer harm”.
B:“Isolate generated deliverables, notify counsel and correct affected outputs with audit trail” is a control for aI-related inaccurate output is not captured as a reportable quality incident, but the audit finding here is aI incident runbooks do not define fallback and business-continuity response. The relevant measure is “Trigger circuit breaker, approved fallback and incident classification based on customer harm”.
C:“Preserve sensor, model, action and operator traces while engaging safety response” is a control for safety incident evidence and physical consequence monitoring are absent, but the audit finding here is aI incident runbooks do not define fallback and business-continuity response. The relevant measure is “Trigger circuit breaker, approved fallback and incident classification based on customer harm”.
Question 368
In an audit of Incident Response Management Specific to AI, managers say this measure is live: Trigger circuit breaker, approved fallback and incident classification based on customer harm. 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 graph and verified rollback/retirement of derived models
- Corrected filings, reviewer approvals and causal analysis
- Time-stamped failover exercise and incident response timeline
- Unified event reconstruction and emergency-stop telemetry
Explanation
Correct Answer
C:Correct. The proposed records—time-stamped failover exercise and incident response timeline—are directly tied to the asserted safeguard and allow its actual operation to be corroborated. The gap being tested is ai incident runbooks do not define fallback and business-continuity response.
Incorrect Answers
A:“Dependency graph and verified rollback/retirement of derived models” would test model incidents do not trace data lineage and affected derivatives, not the present safeguard against aI incident runbooks do not define fallback and business-continuity response. The expected audit evidence is “Time-stamped failover exercise and incident response timeline”.
B:“Corrected filings, reviewer approvals and causal analysis” would test aI-related inaccurate output is not captured as a reportable quality incident, not the present safeguard against aI incident runbooks do not define fallback and business-continuity response. The expected audit evidence is “Time-stamped failover exercise and incident response timeline”.
D:“Unified event reconstruction and emergency-stop telemetry” would test safety incident evidence and physical consequence monitoring are absent, not the present safeguard against aI incident runbooks do not define fallback and business-continuity response. The expected audit evidence is “Time-stamped failover exercise and incident response timeline”.
Question 369
In a review of Incident Response Management Specific to AI, investigators cannot establish the expected control proof: time-stamped failover exercise and incident response timeline. The underlying event was that a fraud model suddenly rejects nearly every payment after a vendor feature outage. Which control weakness is the MOST likely cause requiring focused audit work?
- Model incidents do not trace data lineage and affected derivatives
- AI incident runbooks do not define fallback and business-continuity response
- AI-related inaccurate output is not captured as a reportable quality incident
- Safety incident evidence and physical consequence monitoring are absent
Explanation
Correct Answer
B:Correct. The observed condition specifically indicates ai incident runbooks do not define fallback and business-continuity response. A meaningful audit response is to verify this diagnosis and then recommend trigger circuit breaker, approved fallback and incident classification based on customer harm.
Incorrect Answers
A:“Model incidents do not trace data lineage and affected derivatives” concerns a separate audit weakness, not the missing assurance described by aI incident runbooks do not define fallback and business-continuity response. The correct cause is “AI incident runbooks do not define fallback and business-continuity response”.
C:“AI-related inaccurate output is not captured as a reportable quality incident” concerns a separate audit weakness, not the missing assurance described by aI incident runbooks do not define fallback and business-continuity response. The correct cause is “AI incident runbooks do not define fallback and business-continuity response”.
D:“Safety incident evidence and physical consequence monitoring are absent” concerns a separate audit weakness, not the missing assurance described by aI incident runbooks do not define fallback and business-continuity response. The correct cause is “AI incident runbooks do not define fallback and business-continuity response”.
Question 370
A public chatbot leaks confidential source snippets, but the response team only resets user passwords. 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?
- Disable affected retrieval/tool routes, preserve evidence and assess disclosure scope
- Revoke agent tools, halt transactions and invoke financial incident coordination
- Link model-provider identifiers to affected services and activate targeted containment
- Track trigger conditions, corrective control, owner and verification after closure
Explanation
Correct Answer
A:Correct. Disable affected retrieval/tool routes, preserve evidence and assess disclosure scope is aimed at the actual failure: ai data-exposure playbook lacks containment of model, retrieval and tool paths. The audit should subsequently obtain containment timeline, query logs and privacy incident assessment to test execution, not merely the existence of a policy.
Incorrect Answers
B:“Revoke agent tools, halt transactions and invoke financial incident coordination” is a control for emergency controls lack a tested stop mechanism for automated action, but the audit finding here is aI data-exposure playbook lacks containment of model, retrieval and tool paths. The relevant measure is “Disable affected retrieval/tool routes, preserve evidence and assess disclosure scope”.
C:“Link model-provider identifiers to affected services and activate targeted containment” is a control for supplier incident process lacks AI service dependency mapping, but the audit finding here is aI data-exposure playbook lacks containment of model, retrieval and tool paths. The relevant measure is “Disable affected retrieval/tool routes, preserve evidence and assess disclosure scope”.
D:“Track trigger conditions, corrective control, owner and verification after closure” is a control for root-cause analysis and recurrence prevention are absent from ai incident management, but the audit finding here is aI data-exposure playbook lacks containment of model, retrieval and tool paths. The relevant measure is “Disable affected retrieval/tool routes, preserve evidence and assess disclosure scope”.