Isaca AAIA Practice Test 88
Question 870
An independent AI auditor reviews this case: A health risk model's detailed prediction responses permit inference of protected attributes beyond the intended task. Which TWO complementary elements should the auditor seek: an effective corrective control and the most probative independent evidence? Select TWO answers.
- Evaluate unintended inference from model outputs and minimize observable detail and query access.
- Rate-limit and monitor queries, restrict unnecessary output detail and evaluate extraction risks proportionate to value.
- Inspect pipeline roles, artifact signatures, dataset permissions, approval gates and production change logs.
- Assume suppressing direct database access prevents all leakage of private attributes through an API.
- Inspect output schemas, reconstruction probes, privacy threat model, aggregation and access restrictions.
Explanation
Correct Answers
A:The reason it is necessary is: A model can disclose information statistically through outputs even without returning underlying records verbatim. For this situation: A health risk model's detailed prediction responses permit inference of protected attributes beyond the intended task.
E:The reason it is necessary is: A model can disclose information statistically through outputs even without returning underlying records verbatim. The corrective action is: Evaluate unintended inference from model outputs and minimize observable detail and query access.
Incorrect Answers
B:This option concerns AI model extraction and excessive query access, which is a related but different AI governance concern. In this case: A health risk model's detailed prediction responses permit inference of protected attributes beyond the intended task.
C:This option concerns AI data and model CI/CD separation, which is a related but different AI governance concern. For this situation: A health risk model's detailed prediction responses permit inference of protected attributes beyond the intended task.
D:This attractive shortcut does not resolve the Model inversion and sensitive attribute leakage weakness: Assume suppressing direct database access prevents all leakage of private attributes through an API. For this situation: A health risk model's detailed prediction responses permit inference of protected attributes beyond the intended task.
Question 871
A governance and assurance review uncovers the following: A health risk model's detailed prediction responses permit inference of protected attributes beyond the intended task. What evidence should the auditor examine FIRST to substantiate or challenge the control owner’s claim? Select ONE best answer.
- Assume suppressing direct database access prevents all leakage of private attributes through an API.
- Inspect feature timestamps, label windows, join keys, leakage tests and reconstructed historical decisions.
- Inspect output schemas, reconstruction probes, privacy threat model, aggregation and access restrictions.
- Inspect authentication, rate limits, output granularity, query anomalies and abuse response.
Explanation
Correct Answer
C:The reason it is necessary is: A model can disclose information statistically through outputs even without returning underlying records verbatim. The acceptable verified outcome is: Unintended sensitive inference is bounded and logged access follows approved purpose.
Incorrect Answers
A:This attractive shortcut does not resolve the Model inversion and sensitive attribute leakage weakness: Assume suppressing direct database access prevents all leakage of private attributes through an API. Evaluate unintended inference from model outputs and minimize observable detail and query access.
B:This option concerns Feature-store time leakage in model evaluation, which is a related but different AI governance concern. For this situation: A health risk model's detailed prediction responses permit inference of protected attributes beyond the intended task.
D:This option concerns AI model extraction and excessive query access, which is a related but different AI governance concern. For this situation: A health risk model's detailed prediction responses permit inference of protected attributes beyond the intended task.
Question 872
A risk-based audit engagement examines this operating condition: A health risk model's detailed prediction responses permit inference of protected attributes beyond the intended task. 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.
- Assume suppressing direct database access prevents all leakage of private attributes through an API.
- Requests outside approved privilege and business policy are refused before any sensitive side effect.
- Unintended sensitive inference is bounded and logged access follows approved purpose.
Explanation
Correct Answer
D:The reason it is necessary is: A model can disclose information statistically through outputs even without returning underlying records verbatim. Independent audit evidence includes: Inspect output schemas, reconstruction probes, privacy threat model, aggregation and access restrictions.
Incorrect Answers
A:This option concerns AI data and model CI/CD separation, which is a related but different AI governance concern. The audit must corroborate: Inspect output schemas, reconstruction probes, privacy threat model, aggregation and access restrictions.
B:For this situation: A health risk model's detailed prediction responses permit inference of protected attributes beyond the intended task.
C:This option concerns AI tool misuse and excessive agency, which is a related but different AI governance concern. The decisive principle here is: A model can disclose information statistically through outputs even without returning underlying records verbatim.
Question 873
An audit committee asks for a defensible assessment of this situation: A company releases embedding vectors derived from private documents without validating reconstruction possibilities. What is the MOST appropriate risk-based audit recommendation that addresses the root control weakness? Select ONE best answer.
- Assume suppressing direct database access prevents all leakage of private attributes through an API.
- Separate development, test and production roles with controlled promotion, scoped credentials and audit trails.
- Scope tools to specific identities and approved operations with pre-execution authorization and transaction caps.
- Evaluate unintended inference from model outputs and minimize observable detail and query access.
Explanation
Correct Answer
D:The reason it is necessary is: A model can disclose information statistically through outputs even without returning underlying records verbatim. In this case: A company releases embedding vectors derived from private documents without validating reconstruction possibilities.
Incorrect Answers
A:This attractive shortcut does not resolve the Model inversion and sensitive attribute leakage weakness: Assume suppressing direct database access prevents all leakage of private attributes through an API. In this case: A company releases embedding vectors derived from private documents without validating reconstruction possibilities.
B:This option concerns AI data and model CI/CD separation, which is a related but different AI governance concern. In this case: A company releases embedding vectors derived from private documents without validating reconstruction possibilities.
C:This option concerns AI tool misuse and excessive agency, which is a related but different AI governance concern. In this case: A company releases embedding vectors derived from private documents without validating reconstruction possibilities.
Question 874
During an AI controls assessment, the team learns that a company releases embedding vectors derived from private documents without validating reconstruction possibilities. Which principle BEST explains why the observed practice is a material governance or control concern? Select ONE best answer.
- Repeated API observations can reveal model behavior or sensitive correlations even without access to the weight files.
- A model can disclose information statistically through outputs even without returning underlying records verbatim.
- Model-generated text is untrusted as an authorization basis for real-world privileged actions.
- Assume suppressing direct database access prevents all leakage of private attributes through an API.
Explanation
Correct Answer
B:Independent audit evidence includes: Inspect output schemas, reconstruction probes, privacy threat model, aggregation and access restrictions. In this case: A company releases embedding vectors derived from private documents without validating reconstruction possibilities.
Incorrect Answers
A:This option concerns AI model extraction and excessive query access, which is a related but different AI governance concern. In this case: A company releases embedding vectors derived from private documents without validating reconstruction possibilities.
C:This option concerns AI tool misuse and excessive agency, which is a related but different AI governance concern. For this situation: A company releases embedding vectors derived from private documents without validating reconstruction possibilities.
D:This attractive shortcut does not resolve the Model inversion and sensitive attribute leakage weakness: Assume suppressing direct database access prevents all leakage of private attributes through an API. For this situation: A company releases embedding vectors derived from private documents without validating reconstruction possibilities.
Question 875
An independent AI auditor reviews this case: A company releases embedding vectors derived from private documents without validating reconstruction possibilities. What evidence should the auditor examine FIRST to substantiate or challenge the control owner’s claim? Select ONE best answer.
- Assume suppressing direct database access prevents all leakage of private attributes through an API.
- Inspect signatures, hash attestations, artifact ACLs, dependency inventory and rebuild records.
- Inspect output schemas, reconstruction probes, privacy threat model, aggregation and access restrictions.
- Inspect authentication, rate limits, output granularity, query anomalies and abuse response.
Explanation
Correct Answer
C:In this case: A company releases embedding vectors derived from private documents without validating reconstruction possibilities.
Incorrect Answers
A:For this situation: A company releases embedding vectors derived from private documents without validating reconstruction possibilities.
B:This option concerns Foundation model dependency and artifact tampering, which is a related but different AI governance concern. It does not establish control effectiveness for Model inversion and sensitive attribute leakage.
D:This option concerns AI model extraction and excessive query access, which is a related but different AI governance concern. For this situation: A company releases embedding vectors derived from private documents without validating reconstruction possibilities.
Question 876
A governance and assurance review uncovers the following: A company releases embedding vectors derived from private documents without validating reconstruction possibilities. After management addresses the issue, which result MOST convincingly demonstrates effective remediation? Select ONE best answer.
- Unintended sensitive inference is bounded and logged access follows approved purpose.
- Assume suppressing direct database access prevents all leakage of private attributes through an API.
- Only approved corpus changes influence outputs, and unexpected ranking changes trigger investigation.
- Held-out evaluation uses only information available at the decision time and predicts realistic production performance.
Explanation
Correct Answer
A:This is an observable validation result for Model inversion and sensitive attribute leakage, not just management assurance: Unintended sensitive inference is bounded and logged access follows approved purpose.
Incorrect Answers
B:It misses the key control principle: A model can disclose information statistically through outputs even without returning underlying records verbatim.
C:This option concerns Poisoned embedding and vector retrieval index, which is a related but different AI governance concern. The decisive principle here is: A model can disclose information statistically through outputs even without returning underlying records verbatim.
D:This option concerns Feature-store time leakage in model evaluation, which is a related but different AI governance concern. In this case: A company releases embedding vectors derived from private documents without validating reconstruction possibilities.
Question 877
A risk-based audit engagement examines this operating condition: A customer financial assistant advises a risky unauthorized account transfer and customers begin following it. What is the MOST appropriate risk-based audit recommendation that addresses the root control weakness? Select ONE best answer.
- Treat unsafe guidance as only a normal chatbot quality issue and continue distribution without containment.
- Version and trace training, validation and inference datasets with accountable lineage, hashes and reproducible pipeline metadata.
- Activate an AI-specific incident plan to constrain outputs, preserve evidence, notify responsible owners and restore validated service.
- Quarantine suspicious inputs and models, preserve provenance, revert to validated artifacts and investigate contamination.
Explanation
Correct Answer
C:The reason it is necessary is: AI incidents can harm through decisions and communication even when infrastructure is available. Independent audit evidence includes: Inspect affected prompts and versions, retrieval health, distribution channels, case severity and containment record.
Incorrect Answers
A:This attractive shortcut does not resolve the Response to harmful AI-generated advice weakness: Treat unsafe guidance as only a normal chatbot quality issue and continue distribution without containment.
B: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 Response to harmful AI-generated advice.
D: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 Response to harmful AI-generated advice.
Question 878
An audit committee asks for a defensible assessment of this situation: A customer financial assistant advises a risky unauthorized account transfer and customers begin following it. Which principle BEST explains why the observed practice is a material governance or control concern? Select ONE best answer.
- AI incidents can harm through decisions and communication even when infrastructure is available.
- AI privacy incidents require ordinary breach governance plus model, retrieval and prompt-specific evidence.
- Treat unsafe guidance as only a normal chatbot quality issue and continue distribution without containment.
- Garbage or inconsistent ground-truth labels cap model validity even if the learning algorithm is sophisticated.
Explanation
Correct Answer
A:Independent audit evidence includes: Inspect affected prompts and versions, retrieval health, distribution channels, case severity and containment record. This is the applicable AI audit principle for Response to harmful AI-generated advice: AI incidents can harm through decisions and communication even when infrastructure is available.
Incorrect Answers
B: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 Response to harmful AI-generated advice.
C:This attractive shortcut does not resolve the Response to harmful AI-generated advice weakness: Treat unsafe guidance as only a normal chatbot quality issue and continue distribution without containment. For this situation: A customer financial assistant advises a risky unauthorized account transfer and customers begin following it.
D:This option concerns Label quality and human annotator agreement, which is a related but different AI governance concern. It does not establish control effectiveness for Response to harmful AI-generated advice.
Question 879
During an AI controls assessment, the team learns that a customer financial assistant advises a risky unauthorized account transfer and customers begin following it. What evidence should the auditor examine FIRST to substantiate or challenge the control owner’s claim? Select ONE best answer.
- Inspect ingestion bursts, source identities, dataset hashes, registry links and clean restore evidence.
- Inspect dataset snapshots, feature pipeline commits, record counts, model registry links and training execution manifests.
- Inspect affected prompts and versions, retrieval health, distribution channels, case severity and containment record.
- Treat unsafe guidance as only a normal chatbot quality issue and continue distribution without containment.
Explanation
Correct Answer
C:The reason it is necessary is: AI incidents can harm through decisions and communication even when infrastructure is available. This evidence directly tests the asserted control for Response to harmful AI-generated advice: Inspect affected prompts and versions, retrieval health, distribution channels, case severity and containment record.
Incorrect Answers
A:This option concerns Model poisoning incident response and rollback, which is a related but different AI governance concern. The decisive principle here is: AI incidents can harm through decisions and communication even when infrastructure is available.
B:This option concerns AI training dataset lineage and reproducibility, which is a related but different AI governance concern. The decisive principle here is: AI incidents can harm through decisions and communication even when infrastructure is available.
D:This attractive shortcut does not resolve the Response to harmful AI-generated advice weakness: Treat unsafe guidance as only a normal chatbot quality issue and continue distribution without containment. The auditor should instead examine: Inspect affected prompts and versions, retrieval health, distribution channels, case severity and containment record.
Question 880
An independent AI auditor reviews this case: A customer financial assistant advises a risky unauthorized account transfer and customers begin following it. Which TWO complementary elements should the auditor seek: an effective corrective control and the most probative independent evidence? Select TWO answers.
- Treat unsafe guidance as only a normal chatbot quality issue and continue distribution without containment.
- Activate an AI-specific incident plan to constrain outputs, preserve evidence, notify responsible owners and restore validated service.
- Review label guidelines, inter-rater reliability, adjudication logs, sampled errors and retraining effects.
- Inspect affected prompts and versions, retrieval health, distribution channels, case severity and containment record.
- Version and trace training, validation and inference datasets with accountable lineage, hashes and reproducible pipeline metadata.
Explanation
Correct Answers
B:The reason it is necessary is: AI incidents can harm through decisions and communication even when infrastructure is available. This response fits the risk associated with Response to harmful AI-generated advice: Activate an AI-specific incident plan to constrain outputs, preserve evidence, notify responsible owners and restore validated service.
D:The reason it is necessary is: AI incidents can harm through decisions and communication even when infrastructure is available. For this situation: A customer financial assistant advises a risky unauthorized account transfer and customers begin following it.
Incorrect Answers
A:This attractive shortcut does not resolve the Response to harmful AI-generated advice weakness: Treat unsafe guidance as only a normal chatbot quality issue and continue distribution without containment. Effective closure requires: Harmful advice is contained, affected users receive appropriate remediation and outputs pass regression tests.
C:This option concerns Label quality and human annotator agreement, which is a related but different AI governance concern. The decisive principle here is: AI incidents can harm through decisions and communication even when infrastructure is available.
E:This option concerns AI training dataset lineage and reproducibility, which is a related but different AI governance concern. For this situation: A customer financial assistant advises a risky unauthorized account transfer and customers begin following it.
