Microsoft AB-731 Practice Test 67
Question 660
At BrightPath Media, a bank purchases an AI service without reviewing vendor processing terms. Evidence: Contract data and prompts may be handled under unclear commitments. The business goal is to review third-party AI risks before procurement approval. The operating team needs an ongoing check after the initial rollout. Which post-rollout signal is most important for detecting regression or governance drift?
- Monitor vendor incident notices and material terms changes
- Monitor usage spikes and billing reconciliation delays
- Monitor unresolved complaints and disparate correction times
- Monitor rubber-stamp review and bypassed approvals
Explanation
Correct Answer
A:
Contract data and prompts may be handled under unclear commitments. Monitor vendor incident notices and material terms changes.
Incorrect Answers
B:
“Monitor usage spikes and billing reconciliation delays” monitors this signal: Consumption can rise faster than the annual forecast. This question instead requires an action that can review third-party AI risks before procurement approval. Contract data and prompts may be handled under unclear commitments. Monitor vendor incident notices and material terms changes.
C:
“Monitor unresolved complaints and disparate correction times” monitors this signal: Users may be unable to correct harmful mistakes. This question instead requires an action that can review third-party AI risks before procurement approval. Contract data and prompts may be handled under unclear commitments. Monitor vendor incident notices and material terms changes.
D:
“Monitor rubber-stamp review and bypassed approvals” monitors this signal: Ambiguous policy exceptions can harm individuals. This question instead requires an action that can review third-party AI risks before procurement approval. Contract data and prompts may be handled under unclear commitments. Monitor vendor incident notices and material terms changes.
Question 661
At Lakeshore Commerce, a corporate assistant must answer from frequently updated manuals. Evidence: A model's static training data cannot reliably track those files. The business goal is to use Azure AI Search indexing as a retrieval component where appropriate. Only non-sensitive synthetic data and approved sandbox systems may be used. Which bounded trial with synthetic records best tests the proposed approach?
- Run fictional queries with deliberately distractor paragraphs
- Query fictional clauses by identifier and paraphrased concept
- Try a simulated inventory update with denied and allowed mock roles
- Index mock manuals then ask questions requiring cited passages
Explanation
Correct Answer
D:
A model's static training data cannot reliably track those files. Index mock manuals then ask questions requiring cited passages. Static model knowledge is not a dependable inventory of frequently changing private files. An authorized search index provides retrievable, versioned content when ingestion and permissions are maintained.
Incorrect Answers
A:
“Run fictional queries with deliberately distractor paragraphs” tests a proposed change for the condition: First-stage keyword search has poor top-result precision. This question instead requires an action that can use Azure AI Search indexing as a retrieval component where appropriate. A model's static training data cannot reliably track those files. Index mock manuals then ask questions requiring cited passages. Static model knowledge is not a dependable inventory of frequently changing private files. An authorized search index provides retrievable, versioned content when ingestion and permissions are maintained.
B:
“Query fictional clauses by identifier and paraphrased concept” tests a proposed change for the condition: Vector-only matching risks missing literal identifiers. This question instead requires an action that can use Azure AI Search indexing as a retrieval component where appropriate. A model's static training data cannot reliably track those files. Index mock manuals then ask questions requiring cited passages. Static model knowledge is not a dependable inventory of frequently changing private files. An authorized search index provides retrievable, versioned content when ingestion and permissions are maintained.
C:
“Try a simulated inventory update with denied and allowed mock roles” tests a proposed change for the condition: Reading policy is different from invoking a high-impact connector action. This question instead requires an action that can use Azure AI Search indexing as a retrieval component where appropriate. A model's static training data cannot reliably track those files. Index mock manuals then ask questions requiring cited passages. Static model knowledge is not a dependable inventory of frequently changing private files. An authorized search index provides retrievable, versioned content when ingestion and permissions are maintained.
Question 662
At Blue Ridge Technology, a CFO estimates AI expense from model tokens alone. Evidence: Integration, human review, logging and employee training are omitted. The business goal is to include operational costs beyond the advertised model rate. The proposed solution must respect documented authority and organizational controls. Which approach most directly meets the stated business objective within the approved authority?
- Use synthetic data for safe iteration but validate with governed representative samples
- Use authorized transcription then a grounded summary with uncertainty labels
- Build a multi-year TCO comparison with sensitivity ranges
- Fix ingestion mapping and revalidate grounded outputs
Explanation
Correct Answer
C:
Integration, human review, logging and employee training are omitted. Build a multi-year TCO comparison with sensitivity ranges.
Incorrect Answers
A:
“Use synthetic data for safe iteration but validate with governed representative samples” is a design or implementation response to the condition: Synthetic samples may reproduce omissions and hide rare real-world issues. This question instead requires an action that can include operational costs beyond the advertised model rate. Integration, human review, logging and employee training are omitted. Build a multi-year TCO comparison with sensitivity ranges.
B:
“Use authorized transcription then a grounded summary with uncertainty labels” is a design or implementation response to the condition: Poor transcription can propagate incorrect speaker attribution into summaries. This question instead requires an action that can include operational costs beyond the advertised model rate. Integration, human review, logging and employee training are omitted. Build a multi-year TCO comparison with sensitivity ranges.
D:
“Fix ingestion mapping and revalidate grounded outputs” is a design or implementation response to the condition: Retrieved field meanings changed even though model version is unchanged. This question instead requires an action that can include operational costs beyond the advertised model rate. Integration, human review, logging and employee training are omitted. Build a multi-year TCO comparison with sensitivity ranges.
Question 663
At Atlas Design, a healthcare company must keep some AI workloads in a permitted region. Evidence: Not every model and deployment option exists in all regions. The business goal is to verify availability and residency before approving architecture. The sponsor needs an evidence-based diagnosis rather than a guessed fix. Which initial investigation best identifies the controlling constraint before any solution is chosen?
- Check chosen model, quota, region and data-processing terms
- Measure exact identifier and conceptual query relevance separately
- Identify the app surface, license and actual connected sources
- Identify relevant harm categories and platform filtering limits
Explanation
Correct Answer
A:
Not every model and deployment option exists in all regions. Check chosen model, quota, region and data-processing terms.
Incorrect Answers
B:
“Measure exact identifier and conceptual query relevance separately” is a diagnostic step for the reported condition: Vector-only matching risks missing literal identifiers. This question instead requires an action that can verify availability and residency before approving architecture. Not every model and deployment option exists in all regions. Check chosen model, quota, region and data-processing terms.
C:
“Identify the app surface, license and actual connected sources” is a diagnostic step for the reported condition: The service has not been authorized to access the requested private file. This question instead requires an action that can verify availability and residency before approving architecture. Not every model and deployment option exists in all regions. Check chosen model, quota, region and data-processing terms.
D:
“Identify relevant harm categories and platform filtering limits” is a diagnostic step for the reported condition: The release needs safety checks beyond normal accuracy tests. This question instead requires an action that can verify availability and residency before approving architecture. Not every model and deployment option exists in all regions. Check chosen model, quota, region and data-processing terms.
Question 664
At Coral Bay Systems, an employee chatbot works poorly for screen-reader users. Evidence: Keyboard navigation and alternative text were not tested. The business goal is to include accessibility as an AI release requirement. The operating team needs an ongoing check after the initial rollout. Which post-rollout signal is most important for detecting regression or governance drift?
- Monitor opaque denials and unresolved appeals
- Monitor attempted privilege escalation and approval bypass
- Monitor license utilization and changes to product entitlements
- Monitor accessibility complaints and new interface regressions
Explanation
Correct Answer
D:
Keyboard navigation and alternative text were not tested. Monitor accessibility complaints and new interface regressions.
Incorrect Answers
A:
“Monitor opaque denials and unresolved appeals” monitors this signal: Customers need a comprehensible explanation and correction mechanism. This question instead requires an action that can include accessibility as an AI release requirement. Keyboard navigation and alternative text were not tested. Monitor accessibility complaints and new interface regressions.
B:
“Monitor attempted privilege escalation and approval bypass” monitors this signal: The agent's action scope could exceed its intended advisory role. This question instead requires an action that can include accessibility as an AI release requirement. Keyboard navigation and alternative text were not tested. Monitor accessibility complaints and new interface regressions.
C:
“Monitor license utilization and changes to product entitlements” monitors this signal: Usage patterns vary and some workflows may use included chat capabilities. This question instead requires an action that can include accessibility as an AI release requirement. Keyboard navigation and alternative text were not tested. Monitor accessibility complaints and new interface regressions.
Question 665
At Riverbend Publishing, a warehouse needs to identify visible damage in package photos. Evidence: Text-only signals cannot describe the damage pattern. The business goal is to evaluate image analysis using Vision capabilities within Foundry Tools. Only non-sensitive synthetic data and approved sandbox systems may be used. Which bounded trial with synthetic records best tests the proposed approach?
- Try a synthetic browser client with no privileged token
- Build a fictional five-slide board deck from a sample report
- Test mock package images with scratches and harmless markings
- Process mock invoices with rotated and low-contrast pages
Explanation
Correct Answer
C:
Text-only signals cannot describe the damage pattern. Test mock package images with scratches and harmless markings.
Incorrect Answers
A:
“Try a synthetic browser client with no privileged token” tests a proposed change for the condition: Exposed credentials could be reused for unauthorized inference. This question instead requires an action that can evaluate image analysis using Vision capabilities within Foundry Tools. Text-only signals cannot describe the damage pattern. Test mock package images with scratches and harmless markings.
B:
“Build a fictional five-slide board deck from a sample report” tests a proposed change for the condition: Slides need reviewable editing in the target presentation. This question instead requires an action that can evaluate image analysis using Vision capabilities within Foundry Tools. Text-only signals cannot describe the damage pattern. Test mock package images with scratches and harmless markings.
D:
“Process mock invoices with rotated and low-contrast pages” tests a proposed change for the condition: The original files are scans and require text extraction. This question instead requires an action that can evaluate image analysis using Vision capabilities within Foundry Tools. Text-only signals cannot describe the damage pattern. Test mock package images with scratches and harmless markings.
Question 666
At Willow Creek Finance, a finance assistant answers a query and is also asked to transfer funds. Evidence: Retrieving an account statement does not authorize transactions. The business goal is to separate read-only grounding from high-impact actions. The project owner requires a measurable acceptance criterion before signoff. Which evidence would most convincingly demonstrate that the requirement has been met?
- The assistant informs without executing unauthorized payment actions
- The model cannot ship without required signoffs and rollback plan
- Correct relevant paragraphs are retrieved without losing the heading context
- The returned wording matches the approved source and its citation
Explanation
Correct Answer
A:
Retrieving an account statement does not authorize transactions. The assistant informs without executing unauthorized payment actions.
Incorrect Answers
B:
“The model cannot ship without required signoffs and rollback plan” validates an outcome in a case where the evidence is: There is no documented validation, deployment or monitoring plan. This question instead requires an action that can separate read-only grounding from high-impact actions. Retrieving an account statement does not authorize transactions. The assistant informs without executing unauthorized payment actions.
C:
“Correct relevant paragraphs are retrieved without losing the heading context” validates an outcome in a case where the evidence is: Large irrelevant text consumes context and obscures the answer. This question instead requires an action that can separate read-only grounding from high-impact actions. Retrieving an account statement does not authorize transactions. The assistant informs without executing unauthorized payment actions.
D:
“The returned wording matches the approved source and its citation” validates an outcome in a case where the evidence is: Paraphrasing would change obligations and cannot replace the signed source. This question instead requires an action that can separate read-only grounding from high-impact actions. Retrieving an account statement does not authorize transactions. The assistant informs without executing unauthorized payment actions.
Question 667
At Evergreen Research, a support team wants a first prototype for general email summarization. Evidence: No task-specific labeled training data has yet been collected. The business goal is to use a pretrained model as a measured baseline before customization. The sponsor needs an evidence-based diagnosis rather than a guessed fix. Which initial investigation best identifies the controlling constraint before any solution is chosen?
- Determine whether representative examples already meet acceptance criteria
- Break down outcomes by relevant cohorts and missing data
- Define quality dimensions and disagreements before the review
- Verify sensors, failure labels and class imbalance
Explanation
Correct Answer
A:
No task-specific labeled training data has yet been collected. Determine whether representative examples already meet acceptance criteria.
Incorrect Answers
B:
“Break down outcomes by relevant cohorts and missing data” is a diagnostic step for the reported condition: Training examples omit important customer contexts. This question instead requires an action that can use a pretrained model as a measured baseline before customization. No task-specific labeled training data has yet been collected. Determine whether representative examples already meet acceptance criteria.
C:
“Define quality dimensions and disagreements before the review” is a diagnostic step for the reported condition: No documented rubric or blind comparison exists. This question instead requires an action that can use a pretrained model as a measured baseline before customization. No task-specific labeled training data has yet been collected. Determine whether representative examples already meet acceptance criteria.
D:
“Verify sensors, failure labels and class imbalance” is a diagnostic step for the reported condition: Threshold rules miss complex multi-variable patterns. This question instead requires an action that can use a pretrained model as a measured baseline before customization. No task-specific labeled training data has yet been collected. Determine whether representative examples already meet acceptance criteria.
Question 668
At Atlas Design, an executive requests a vendor market scan and a forecast from internal sales data. Evidence: External research and numerical modeling are distinct activities. The business goal is to assign specialized Copilot capabilities to appropriate subtasks. The project owner requires a measurable acceptance criterion before signoff. Which evidence would most convincingly demonstrate that the requirement has been met?
- Each conclusion is sourced to the correct evidence stream
- Unseen-case format gains outweigh training and hosting costs
- The answer acknowledges disagreement rather than implying consensus
- Both editors see consistent reviewed updates without overbroad sharing
Explanation
Correct Answer
A:
External research and numerical modeling are distinct activities. Each conclusion is sourced to the correct evidence stream.
Incorrect Answers
B:
“Unseen-case format gains outweigh training and hosting costs” validates an outcome in a case where the evidence is: An existing prompt baseline still violates approved template constraints. This question instead requires an action that can assign specialized Copilot capabilities to appropriate subtasks. External research and numerical modeling are distinct activities. Each conclusion is sourced to the correct evidence stream.
C:
“The answer acknowledges disagreement rather than implying consensus” validates an outcome in a case where the evidence is: The agent draws from a small biased sample of industry articles. This question instead requires an action that can assign specialized Copilot capabilities to appropriate subtasks. External research and numerical modeling are distinct activities. Each conclusion is sourced to the correct evidence stream.
D:
“Both editors see consistent reviewed updates without overbroad sharing” validates an outcome in a case where the evidence is: Parallel copies cause conflicting versions of one plan. This question instead requires an action that can assign specialized Copilot capabilities to appropriate subtasks. External research and numerical modeling are distinct activities. Each conclusion is sourced to the correct evidence stream.
Question 669
At Cedar Analytics, a multinational team asks for AI-generated slide content accessible to all employees. Evidence: Generated visuals may omit alt text or use low-contrast design. The business goal is to pair Copilot drafting with accessibility review. The project owner requires a measurable acceptance criterion before signoff. Which evidence would most convincingly demonstrate that the requirement has been met?
- Requests route to intended topics without guessing sensitive details
- Final deck remains understandable using keyboard and screen reader
- Unauthorized clients cannot invoke privileged endpoints
- Sensitive repositories remain inaccessible to unknown agents
Explanation
Correct Answer
B:
Generated visuals may omit alt text or use low-contrast design. Final deck remains understandable using keyboard and screen reader.
Incorrect Answers
A:
“Requests route to intended topics without guessing sensitive details” validates an outcome in a case where the evidence is: Different user intents need separate approved answer paths. This question instead requires an action that can pair Copilot drafting with accessibility review. Generated visuals may omit alt text or use low-contrast design. Final deck remains understandable using keyboard and screen reader.
C:
“Unauthorized clients cannot invoke privileged endpoints” validates an outcome in a case where the evidence is: Exposed credentials could be reused for unauthorized inference. This question instead requires an action that can pair Copilot drafting with accessibility review. Generated visuals may omit alt text or use low-contrast design. Final deck remains understandable using keyboard and screen reader.
D:
“Sensitive repositories remain inaccessible to unknown agents” validates an outcome in a case where the evidence is: Publisher approval and data access are uncertain. This question instead requires an action that can pair Copilot drafting with accessibility review. Generated visuals may omit alt text or use low-contrast design. Final deck remains understandable using keyboard and screen reader.
Question 670
At Elmwood Banking, an audit department queries a policy assistant that cites a superseded manual. Evidence: The knowledge index was last refreshed before the recent policy revision. The business goal is to maintain fresh retrieval sources rather than changing model weights. Only non-sensitive synthetic data and approved sandbox systems may be used. Which bounded trial with synthetic records best tests the proposed approach?
- Simulate a bad source date and verify sends are blocked
- Test retrieval of mock credentials by authorized and denied roles
- Run a mock benchmark with intentionally overlapping cases
- Replace a fictional policy version and rerun the same question
Explanation
Correct Answer
D:
The knowledge index was last refreshed before the recent policy revision. Replace a fictional policy version and rerun the same question. A policy that changes after indexing will not be reflected in retrieved passages until the source is refreshed. Track source versions, ingestion success and index freshness instead of retraining the model for factual updates.
Incorrect Answers
A:
“Simulate a bad source date and verify sends are blocked” tests a proposed change for the condition: One incorrect dispatch date could be repeated to thousands of customers. This question instead requires an action that can maintain fresh retrieval sources rather than changing model weights. The knowledge index was last refreshed before the recent policy revision. Replace a fictional policy version and rerun the same question. A policy that changes after indexing will not be reflected in retrieved passages until the source is refreshed. Track source versions, ingestion success and index freshness instead of retraining the model for factual updates.
B:
“Test retrieval of mock credentials by authorized and denied roles” tests a proposed change for the condition: The credentials enable unauthorized paid inference requests. This question instead requires an action that can maintain fresh retrieval sources rather than changing model weights. The knowledge index was last refreshed before the recent policy revision. Replace a fictional policy version and rerun the same question. A policy that changes after indexing will not be reflected in retrieved passages until the source is refreshed. Track source versions, ingestion success and index freshness instead of retraining the model for factual updates.
C:
“Run a mock benchmark with intentionally overlapping cases” tests a proposed change for the condition: The same customer records appear in tuning and testing data. This question instead requires an action that can maintain fresh retrieval sources rather than changing model weights. The knowledge index was last refreshed before the recent policy revision. Replace a fictional policy version and rerun the same question. A policy that changes after indexing will not be reflected in retrieved passages until the source is refreshed. Track source versions, ingestion success and index freshness instead of retraining the model for factual updates.