ISACA Advanced in AI Audit AAIA Practice Test 13
Question 120
In a review of Privacy and Data Governance Programs, investigators cannot establish the expected control proof: end-to-end deletion trace from source to vector index and exception ledger. The underlying event was that a deletion request is completed in the CRM but old customer examples remain in retrieval indexes. Which control weakness is the MOST likely cause requiring focused audit work?
- AI data transfers are missing location and processor transparency
- Retention defaults violate purpose limitation and approved data minimization
- Subject-rights workflow excludes derived AI data stores and vector copies
- Reidentification risk was not tested for the actual release context
Explanation
Correct Answer
C:Correct. The observed condition specifically indicates subject-rights workflow excludes derived ai data stores and vector copies. A meaningful audit response is to verify this diagnosis and then recommend extend deletion propagation to indexes, caches, fine-tune stores and backups as applicable.
Incorrect Answers
A:“AI data transfers are missing location and processor transparency” concerns a separate audit weakness, not the missing assurance described by subject-rights workflow excludes derived AI data stores and vector copies. The correct cause is “Subject-rights workflow excludes derived AI data stores and vector copies”.
B:“Retention defaults violate purpose limitation and approved data minimization” concerns a separate audit weakness, not the missing assurance described by subject-rights workflow excludes derived AI data stores and vector copies. The correct cause is “Subject-rights workflow excludes derived AI data stores and vector copies”.
D:“Reidentification risk was not tested for the actual release context” concerns a separate audit weakness, not the missing assurance described by subject-rights workflow excludes derived AI data stores and vector copies. The correct cause is “Subject-rights workflow excludes derived AI data stores and vector copies”.
Question 121
Dataset stewards approve completeness but cannot identify which fields were imputed or transformed. 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?
- Evaluate memorization, membership inference and release utility before sharing
- Capture versioned transformation logic, imputation decisions and field-level lineage
- Require documented lineage, rights basis and permitted-use checks before ingestion
- Sanitize and segregate prompt traces with restricted access and retention
Explanation
Correct Answer
B:Correct. Capture versioned transformation logic, imputation decisions and field-level lineage is aimed at the actual failure: lineage controls omit preprocessing operations essential to auditability. The audit should subsequently obtain reproducible pipeline manifests and source-to-feature trace samples to test execution, not merely the existence of a policy.
Incorrect Answers
A:“Evaluate memorization, membership inference and release utility before sharing” is a control for synthetic data was assumed private without membership-inference or disclosure tests, but the audit finding here is lineage controls omit preprocessing operations essential to auditability. The relevant measure is “Capture versioned transformation logic, imputation decisions and field-level lineage”.
C:“Require documented lineage, rights basis and permitted-use checks before ingestion” is a control for data provenance and lawful-use assurance are not established, but the audit finding here is lineage controls omit preprocessing operations essential to auditability. The relevant measure is “Capture versioned transformation logic, imputation decisions and field-level lineage”.
D:“Sanitize and segregate prompt traces with restricted access and retention” is a control for observability logs bypass personal-data classification and handling controls, but the audit finding here is lineage controls omit preprocessing operations essential to auditability. The relevant measure is “Capture versioned transformation logic, imputation decisions and field-level lineage”.
Question 122
In an audit of Privacy and Data Governance Programs, managers say this measure is live: Capture versioned transformation logic, imputation decisions and field-level lineage. 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?
- Reproducible pipeline manifests and source-to-feature trace samples
- Attack evaluation reports, privacy thresholds and restricted release approvals
- Data source records, licensing terms and approval evidence for each training corpus
- Log pipeline samples, redaction tests and access reviews
Explanation
Correct Answer
A:Correct. The proposed records—reproducible pipeline manifests and source-to-feature trace samples—are directly tied to the asserted safeguard and allow its actual operation to be corroborated. The gap being tested is lineage controls omit preprocessing operations essential to auditability.
Incorrect Answers
B:“Attack evaluation reports, privacy thresholds and restricted release approvals” would test synthetic data was assumed private without membership-inference or disclosure tests, not the present safeguard against lineage controls omit preprocessing operations essential to auditability. The expected audit evidence is “Reproducible pipeline manifests and source-to-feature trace samples”.
C:“Data source records, licensing terms and approval evidence for each training corpus” would test data provenance and lawful-use assurance are not established, not the present safeguard against lineage controls omit preprocessing operations essential to auditability. The expected audit evidence is “Reproducible pipeline manifests and source-to-feature trace samples”.
D:“Log pipeline samples, redaction tests and access reviews” would test observability logs bypass personal-data classification and handling controls, not the present safeguard against lineage controls omit preprocessing operations essential to auditability. The expected audit evidence is “Reproducible pipeline manifests and source-to-feature trace samples”.
Question 123
In a review of Privacy and Data Governance Programs, investigators cannot establish the expected control proof: reproducible pipeline manifests and source-to-feature trace samples. The underlying event was that dataset stewards approve completeness but cannot identify which fields were imputed or transformed. Which control weakness is the MOST likely cause requiring focused audit work?
- Synthetic data was assumed private without membership-inference or disclosure tests
- Data provenance and lawful-use assurance are not established
- Observability logs bypass personal-data classification and handling controls
- Lineage controls omit preprocessing operations essential to auditability
Explanation
Correct Answer
D:Correct. The observed condition specifically indicates lineage controls omit preprocessing operations essential to auditability. A meaningful audit response is to verify this diagnosis and then recommend capture versioned transformation logic, imputation decisions and field-level lineage.
Incorrect Answers
A:“Synthetic data was assumed private without membership-inference or disclosure tests” concerns a separate audit weakness, not the missing assurance described by lineage controls omit preprocessing operations essential to auditability. The correct cause is “Lineage controls omit preprocessing operations essential to auditability”.
B:“Data provenance and lawful-use assurance are not established” concerns a separate audit weakness, not the missing assurance described by lineage controls omit preprocessing operations essential to auditability. The correct cause is “Lineage controls omit preprocessing operations essential to auditability”.
C:“Observability logs bypass personal-data classification and handling controls” concerns a separate audit weakness, not the missing assurance described by lineage controls omit preprocessing operations essential to auditability. The correct cause is “Lineage controls omit preprocessing operations essential to auditability”.
Question 124
A vendor transfers inference logs across borders without specifying subprocessors. 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?
- Test extraction risk and restrict sensitive content in model training and output handling
- Enforce least-privilege project entitlements and scoped deidentified training views
- Inventory processing locations, subprocessors and cross-border transfer safeguards
- Extend deletion propagation to indexes, caches, fine-tune stores and backups as applicable
Explanation
Correct Answer
C:Correct. Inventory processing locations, subprocessors and cross-border transfer safeguards is aimed at the actual failure: ai data transfers are missing location and processor transparency. The audit should subsequently obtain verified processor list, hosting configuration and applicable transfer assessments to test execution, not merely the existence of a policy.
Incorrect Answers
A:“Test extraction risk and restrict sensitive content in model training and output handling” is a control for training memorization and output disclosure controls were unassessed, but the audit finding here is aI data transfers are missing location and processor transparency. The relevant measure is “Inventory processing locations, subprocessors and cross-border transfer safeguards”.
B:“Enforce least-privilege project entitlements and scoped deidentified training views” is a control for dataset access is broad and lacks purpose-based authorization, but the audit finding here is aI data transfers are missing location and processor transparency. The relevant measure is “Inventory processing locations, subprocessors and cross-border transfer safeguards”.
D:“Extend deletion propagation to indexes, caches, fine-tune stores and backups as applicable” is a control for subject-rights workflow excludes derived AI data stores and vector copies, but the audit finding here is aI data transfers are missing location and processor transparency. The relevant measure is “Inventory processing locations, subprocessors and cross-border transfer safeguards”.
Question 125
In an audit of Privacy and Data Governance Programs, managers say this measure is live: Inventory processing locations, subprocessors and cross-border transfer safeguards. 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?
- Memorization probe results, content filters and reviewed training-data exclusions
- Verified processor list, hosting configuration and applicable transfer assessments
- Access grants, denied cross-project query tests and periodic privilege recertifications
- End-to-end deletion trace from source to vector index and exception ledger
Explanation
Correct Answer
B:Correct. The proposed records—verified processor list, hosting configuration and applicable transfer assessments—are directly tied to the asserted safeguard and allow its actual operation to be corroborated. The gap being tested is ai data transfers are missing location and processor transparency.
Incorrect Answers
A:“Memorization probe results, content filters and reviewed training-data exclusions” would test training memorization and output disclosure controls were unassessed, not the present safeguard against aI data transfers are missing location and processor transparency. The expected audit evidence is “Verified processor list, hosting configuration and applicable transfer assessments”.
C:“Access grants, denied cross-project query tests and periodic privilege recertifications” would test dataset access is broad and lacks purpose-based authorization, not the present safeguard against aI data transfers are missing location and processor transparency. The expected audit evidence is “Verified processor list, hosting configuration and applicable transfer assessments”.
D:“End-to-end deletion trace from source to vector index and exception ledger” would test subject-rights workflow excludes derived AI data stores and vector copies, not the present safeguard against aI data transfers are missing location and processor transparency. The expected audit evidence is “Verified processor list, hosting configuration and applicable transfer assessments”.
Question 126
In a review of Privacy and Data Governance Programs, investigators cannot establish the expected control proof: verified processor list, hosting configuration and applicable transfer assessments. The underlying event was that a vendor transfers inference logs across borders without specifying subprocessors. Which control weakness is the MOST likely cause requiring focused audit work?
- AI data transfers are missing location and processor transparency
- Training memorization and output disclosure controls were unassessed
- Dataset access is broad and lacks purpose-based authorization
- Subject-rights workflow excludes derived AI data stores and vector copies
Explanation
Correct Answer
A:Correct. The observed condition specifically indicates ai data transfers are missing location and processor transparency. A meaningful audit response is to verify this diagnosis and then recommend inventory processing locations, subprocessors and cross-border transfer safeguards.
Incorrect Answers
B:“Training memorization and output disclosure controls were unassessed” concerns a separate audit weakness, not the missing assurance described by aI data transfers are missing location and processor transparency. The correct cause is “AI data transfers are missing location and processor transparency”.
C:“Dataset access is broad and lacks purpose-based authorization” concerns a separate audit weakness, not the missing assurance described by aI data transfers are missing location and processor transparency. The correct cause is “AI data transfers are missing location and processor transparency”.
D:“Subject-rights workflow excludes derived AI data stores and vector copies” concerns a separate audit weakness, not the missing assurance described by aI data transfers are missing location and processor transparency. The correct cause is “AI data transfers are missing location and processor transparency”.
Question 127
A synthetic dataset preserves rare patient records so closely that privacy attacks recover source data. 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 category-specific prompt retention limits, redaction and verified deletion paths
- Conduct purpose/compatibility and privacy impact review before reuse
- Capture versioned transformation logic, imputation decisions and field-level lineage
- Evaluate memorization, membership inference and release utility before sharing
Explanation
Correct Answer
D:Correct. Evaluate memorization, membership inference and release utility before sharing is aimed at the actual failure: synthetic data was assumed private without membership-inference or disclosure tests. The audit should subsequently obtain attack evaluation reports, privacy thresholds and restricted release approvals to test execution, not merely the existence of a policy.
Incorrect Answers
A:“Apply category-specific prompt retention limits, redaction and verified deletion paths” is a control for retention defaults violate purpose limitation and approved data minimization, but the audit finding here is synthetic data was assumed private without membership-inference or disclosure tests. The relevant measure is “Evaluate memorization, membership inference and release utility before sharing”.
B:“Conduct purpose/compatibility and privacy impact review before reuse” is a control for secondary AI processing is not evaluated against the original lawful purpose, but the audit finding here is synthetic data was assumed private without membership-inference or disclosure tests. The relevant measure is “Evaluate memorization, membership inference and release utility before sharing”.
C:“Capture versioned transformation logic, imputation decisions and field-level lineage” is a control for lineage controls omit preprocessing operations essential to auditability, but the audit finding here is synthetic data was assumed private without membership-inference or disclosure tests. The relevant measure is “Evaluate memorization, membership inference and release utility before sharing”.
Question 128
In an audit of Privacy and Data Governance Programs, managers say this measure is live: Evaluate memorization, membership inference and release utility before sharing. 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?
- Configured TTL, deletion test logs and contractual retention commitments
- Updated processing register, legal assessment and approved purpose mapping
- Attack evaluation reports, privacy thresholds and restricted release approvals
- Reproducible pipeline manifests and source-to-feature trace samples
Explanation
Correct Answer
C:Correct. The proposed records—attack evaluation reports, privacy thresholds and restricted release approvals—are directly tied to the asserted safeguard and allow its actual operation to be corroborated. The gap being tested is synthetic data was assumed private without membership-inference or disclosure tests.
Incorrect Answers
A:“Configured TTL, deletion test logs and contractual retention commitments” would test retention defaults violate purpose limitation and approved data minimization, not the present safeguard against synthetic data was assumed private without membership-inference or disclosure tests. The expected audit evidence is “Attack evaluation reports, privacy thresholds and restricted release approvals”.
B:“Updated processing register, legal assessment and approved purpose mapping” would test secondary AI processing is not evaluated against the original lawful purpose, not the present safeguard against synthetic data was assumed private without membership-inference or disclosure tests. The expected audit evidence is “Attack evaluation reports, privacy thresholds and restricted release approvals”.
D:“Reproducible pipeline manifests and source-to-feature trace samples” would test lineage controls omit preprocessing operations essential to auditability, not the present safeguard against synthetic data was assumed private without membership-inference or disclosure tests. The expected audit evidence is “Attack evaluation reports, privacy thresholds and restricted release approvals”.
Question 129
In a review of Privacy and Data Governance Programs, investigators cannot establish the expected control proof: attack evaluation reports, privacy thresholds and restricted release approvals. The underlying event was that a synthetic dataset preserves rare patient records so closely that privacy attacks recover source data. Which control weakness is the MOST likely cause requiring focused audit work?
- Retention defaults violate purpose limitation and approved data minimization
- Synthetic data was assumed private without membership-inference or disclosure tests
- Secondary AI processing is not evaluated against the original lawful purpose
- Lineage controls omit preprocessing operations essential to auditability
Explanation
Correct Answer
B:Correct. The observed condition specifically indicates synthetic data was assumed private without membership-inference or disclosure tests. A meaningful audit response is to verify this diagnosis and then recommend evaluate memorization, membership inference and release utility before sharing.
Incorrect Answers
A:“Retention defaults violate purpose limitation and approved data minimization” concerns a separate audit weakness, not the missing assurance described by synthetic data was assumed private without membership-inference or disclosure tests. The correct cause is “Synthetic data was assumed private without membership-inference or disclosure tests”.
C:“Secondary AI processing is not evaluated against the original lawful purpose” concerns a separate audit weakness, not the missing assurance described by synthetic data was assumed private without membership-inference or disclosure tests. The correct cause is “Synthetic data was assumed private without membership-inference or disclosure tests”.
D:“Lineage controls omit preprocessing operations essential to auditability” concerns a separate audit weakness, not the missing assurance described by synthetic data was assumed private without membership-inference or disclosure tests. The correct cause is “Synthetic data was assumed private without membership-inference or disclosure tests”.
Question 130
A model output accidentally reveals confidential source text verbatim from training. 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?
- Test extraction risk and restrict sensitive content in model training and output handling
- Require documented lineage, rights basis and permitted-use checks before ingestion
- Perform linkage attacks and set release safeguards appropriate to residual identifiability
- Inventory processing locations, subprocessors and cross-border transfer safeguards
Explanation
Correct Answer
A:Correct. Test extraction risk and restrict sensitive content in model training and output handling is aimed at the actual failure: training memorization and output disclosure controls were unassessed. The audit should subsequently obtain memorization probe results, content filters and reviewed training-data exclusions to test execution, not merely the existence of a policy.
Incorrect Answers
B:“Require documented lineage, rights basis and permitted-use checks before ingestion” is a control for data provenance and lawful-use assurance are not established, but the audit finding here is training memorization and output disclosure controls were unassessed. The relevant measure is “Test extraction risk and restrict sensitive content in model training and output handling”.
C:“Perform linkage attacks and set release safeguards appropriate to residual identifiability” is a control for reidentification risk was not tested for the actual release context, but the audit finding here is training memorization and output disclosure controls were unassessed. The relevant measure is “Test extraction risk and restrict sensitive content in model training and output handling”.
D:“Inventory processing locations, subprocessors and cross-border transfer safeguards” is a control for aI data transfers are missing location and processor transparency, but the audit finding here is training memorization and output disclosure controls were unassessed. The relevant measure is “Test extraction risk and restrict sensitive content in model training and output handling”.