Microsoft AB-731 Practice Test 71
Question 700
At Harborline Products, a public chatbot prototype occasionally outputs unsafe or abusive language. Evidence: The release needs safety checks beyond normal accuracy tests. The business goal is to integrate suitable content safety screening into deployment criteria. The proposed solution must respect documented authority and organizational controls. Which approach most directly meets the stated business objective within the approved authority?
- Set content filtering and escalations appropriate to the audience
- Use PowerPoint Copilot to draft slides then review every claim
- Block release until model and application mitigations pass tests
- Create a held-out evaluation dataset and acceptance thresholds
Explanation
Correct Answer
A:
The release needs safety checks beyond normal accuracy tests. Set content filtering and escalations appropriate to the audience. Conventional task accuracy does not measure unsafe content, tool misuse or privacy leakage. Add appropriate adversarial and policy tests as a separate release gate.
Incorrect Answers
B:
“Use PowerPoint Copilot to draft slides then review every claim” is a design or implementation response to the condition: Slides need reviewable editing in the target presentation. This question instead requires an action that can integrate suitable content safety screening into deployment criteria. The release needs safety checks beyond normal accuracy tests. Set content filtering and escalations appropriate to the audience. Conventional task accuracy does not measure unsafe content, tool misuse or privacy leakage. Add appropriate adversarial and policy tests as a separate release gate.
C:
“Block release until model and application mitigations pass tests” is a design or implementation response to the condition: Performance alone does not demonstrate responsible deployment. This question instead requires an action that can integrate suitable content safety screening into deployment criteria. The release needs safety checks beyond normal accuracy tests. Set content filtering and escalations appropriate to the audience. Conventional task accuracy does not measure unsafe content, tool misuse or privacy leakage. Add appropriate adversarial and policy tests as a separate release gate.
D:
“Create a held-out evaluation dataset and acceptance thresholds” is a design or implementation response to the condition: The published scores do not measure the company's actual document tasks. This question instead requires an action that can integrate suitable content safety screening into deployment criteria. The release needs safety checks beyond normal accuracy tests. Set content filtering and escalations appropriate to the audience. Conventional task accuracy does not measure unsafe content, tool misuse or privacy leakage. Add appropriate adversarial and policy tests as a separate release gate.
Question 701
At Crownridge Travel, a strategy manager wants a one-page proposal based on an approved memo. Evidence: The task needs a working document with source-grounded language. The business goal is to use Word-integrated Copilot for drafting with human review. The proposed solution must respect documented authority and organizational controls. Which approach most directly meets the stated business objective within the approved authority?
- Use safe prompt hierarchy, sanitization and authorized tool controls
- Draft inside Word from the authorized memo and review commitments
- Deploy a prebuilt option if it meets scope and licensing requirements
- Define workflow instructions independent of unsupported model toggles
Explanation
Correct Answer
B:
The task needs a working document with source-grounded language. Draft inside Word from the authorized memo and review commitments.
Incorrect Answers
A:
“Use safe prompt hierarchy, sanitization and authorized tool controls” is a design or implementation response to the condition: The document includes text telling the assistant to export hidden records. This question instead requires an action that can use Word-integrated Copilot for drafting with human review. The task needs a working document with source-grounded language. Draft inside Word from the authorized memo and review commitments.
C:
“Deploy a prebuilt option if it meets scope and licensing requirements” is a design or implementation response to the condition: A prebuilt agent may already support search without bespoke code. This question instead requires an action that can use Word-integrated Copilot for drafting with human review. The task needs a working document with source-grounded language. Draft inside Word from the authorized memo and review commitments.
D:
“Define workflow instructions independent of unsupported model toggles” is a design or implementation response to the condition: The supported surface and administrative policies govern model availability. This question instead requires an action that can use Word-integrated Copilot for drafting with human review. The task needs a working document with source-grounded language. Draft inside Word from the authorized memo and review commitments.
Question 702
At Prairie Health, a business user assumes every Copilot agent exposes identical model settings. Evidence: The supported surface and administrative policies govern model availability. The business goal is to use only actually available model choices and document limits. Only non-sensitive synthetic data and approved sandbox systems may be used. Which bounded trial with synthetic records best tests the proposed approach?
- Prototype a simulated approval step without production writes
- Test synthetic product photos paired with conflicting notes
- Compare mock outputs using sanctioned product settings
- Test a fictional department network with hidden membership
Explanation
Correct Answer
C:
The supported surface and administrative policies govern model availability. Compare mock outputs using sanctioned product settings.
Incorrect Answers
A:
“Prototype a simulated approval step without production writes” tests a proposed change for the condition: Base Copilot chat cannot directly implement every business-specific action. This question instead requires an action that can use only actually available model choices and document limits. The supported surface and administrative policies govern model availability. Compare mock outputs using sanctioned product settings.
B:
“Test synthetic product photos paired with conflicting notes” tests a proposed change for the condition: Text-only models cannot inspect image evidence. This question instead requires an action that can use only actually available model choices and document limits. The supported surface and administrative policies govern model availability. Compare mock outputs using sanctioned product settings.
D:
“Test a fictional department network with hidden membership” tests a proposed change for the condition: Graph relationships may expose sensitive work patterns and restricted files. This question instead requires an action that can use only actually available model choices and document limits. The supported surface and administrative policies govern model availability. Compare mock outputs using sanctioned product settings.
Question 703
At Northlake Services, an HR knowledge bot exposes sensitive benefits documents to a general employee. Evidence: Index query results lack caller-specific authorization filters. The business goal is to enforce document-level authorization in retrieval. Only non-sensitive synthetic data and approved sandbox systems may be used. Which bounded trial with synthetic records best tests the proposed approach?
- Test synthetic product photos paired with conflicting notes
- Create a sample proposal from a fictional policy note
- Compare a no-retrieval baseline with an indexed policy pilot
- Test mock restricted files under permitted and denied identities
Explanation
Correct Answer
D:
Index query results lack caller-specific authorization filters. Test mock restricted files under permitted and denied identities. A search index that ignores caller identity can leak restricted snippets before generation. Apply document-level authorization during candidate retrieval and verify denial paths with test users.
Incorrect Answers
A:
“Test synthetic product photos paired with conflicting notes” tests a proposed change for the condition: Text-only models cannot inspect image evidence. This question instead requires an action that can enforce document-level authorization in retrieval. Index query results lack caller-specific authorization filters. Test mock restricted files under permitted and denied identities. A search index that ignores caller identity can leak restricted snippets before generation. Apply document-level authorization during candidate retrieval and verify denial paths with test users.
B:
“Create a sample proposal from a fictional policy note” tests a proposed change for the condition: The task needs a working document with source-grounded language. This question instead requires an action that can enforce document-level authorization in retrieval. Index query results lack caller-specific authorization filters. Test mock restricted files under permitted and denied identities. A search index that ignores caller identity can leak restricted snippets before generation. Apply document-level authorization during candidate retrieval and verify denial paths with test users.
C:
“Compare a no-retrieval baseline with an indexed policy pilot” tests a proposed change for the condition: Fresh facts and style behavior are separate concerns. This question instead requires an action that can enforce document-level authorization in retrieval. Index query results lack caller-specific authorization filters. Test mock restricted files under permitted and denied identities. A search index that ignores caller identity can leak restricted snippets before generation. Apply document-level authorization during candidate retrieval and verify denial paths with test users.
Question 704
At Cedar Analytics, a team embeds Foundry API secrets in customer-facing browser code. Evidence: Exposed credentials could be reused for unauthorized inference. The business goal is to protect model endpoints with least privilege and managed credentials. 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 which connected entities and files the analyst can access
- Record token volumes, retries and actual chosen billing meters
- Compare current failure patterns against baseline benchmarks
- Inspect where tokens are stored and which callers can reach the endpoint
Explanation
Correct Answer
D:
Exposed credentials could be reused for unauthorized inference. Inspect where tokens are stored and which callers can reach the endpoint.
Incorrect Answers
A:
“Determine which connected entities and files the analyst can access” is a diagnostic step for the reported condition: People, files and collaboration events have permissioned relationships. This question instead requires an action that can protect model endpoints with least privilege and managed credentials. Exposed credentials could be reused for unauthorized inference. Inspect where tokens are stored and which callers can reach the endpoint.
B:
“Record token volumes, retries and actual chosen billing meters” is a diagnostic step for the reported condition: Input and output tokens and deployment type all affect billing. This question instead requires an action that can protect model endpoints with least privilege and managed credentials. Exposed credentials could be reused for unauthorized inference. Inspect where tokens are stored and which callers can reach the endpoint.
C:
“Compare current failure patterns against baseline benchmarks” is a diagnostic step for the reported condition: New request types were absent from evaluation data. This question instead requires an action that can protect model endpoints with least privilege and managed credentials. Exposed credentials could be reused for unauthorized inference. Inspect where tokens are stored and which callers can reach the endpoint.
Question 705
At Coral Bay Systems, a utility customer asks a bot to negotiate a nonstandard billing adjustment. Evidence: The request is beyond published policy and automated authority. The business goal is to provide human escalation from governed agents. 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 connector outages, leakage and stale records
- Monitor drift, safety issues and rollback thresholds
- Monitor dropped escalations and unauthorized concession promises
- Monitor repeated single-source bias and unsupported certainty
Explanation
Correct Answer
C:
The request is beyond published policy and automated authority. Monitor dropped escalations and unauthorized concession promises.
Incorrect Answers
A:
“Monitor connector outages, leakage and stale records” monitors this signal: The data is not necessarily available to Microsoft 365 Copilot without configuration. This question instead requires an action that can provide human escalation from governed agents. The request is beyond published policy and automated authority. Monitor dropped escalations and unauthorized concession promises.
B:
“Monitor drift, safety issues and rollback thresholds” monitors this signal: New request types were absent from evaluation data. This question instead requires an action that can provide human escalation from governed agents. The request is beyond published policy and automated authority. Monitor dropped escalations and unauthorized concession promises.
D:
“Monitor repeated single-source bias and unsupported certainty” monitors this signal: The agent draws from a small biased sample of industry articles. This question instead requires an action that can provide human escalation from governed agents. The request is beyond published policy and automated authority. Monitor dropped escalations and unauthorized concession promises.
Question 706
At Summit Manufacturing, a startup proposes training a foundation model from scratch for FAQ answers. Evidence: The available budget and data are far smaller than required for foundation training. The business goal is to prefer evaluating pretrained hosted models when they meet the need. The proposed solution must respect documented authority and organizational controls. Which approach most directly meets the stated business objective within the approved authority?
- Review generated content with accessibility tools and human testers
- Set content filtering and escalations appropriate to the audience
- Test a replacement version with rollback readiness
- Test a suitable existing model before considering custom training
Explanation
Correct Answer
D:
The available budget and data are far smaller than required for foundation training. Test a suitable existing model before considering custom training.
Incorrect Answers
A:
“Review generated content with accessibility tools and human testers” tests a proposed change for the condition: Generated visuals may omit alt text or use low-contrast design. This question instead requires an action that can prefer evaluating pretrained hosted models when they meet the need. The available budget and data are far smaller than required for foundation training. Test a suitable existing model before considering custom training.
B:
“Set content filtering and escalations appropriate to the audience” is a design or implementation response to the condition: The release needs safety checks beyond normal accuracy tests. This question instead requires an action that can prefer evaluating pretrained hosted models when they meet the need. The available budget and data are far smaller than required for foundation training. Test a suitable existing model before considering custom training.
C:
“Test a replacement version with rollback readiness” is a design or implementation response to the condition: The successor may change behavior and cost. This question instead requires an action that can prefer evaluating pretrained hosted models when they meet the need. The available budget and data are far smaller than required for foundation training. Test a suitable existing model before considering custom training.
Question 707
At Sapphire Mobility, a team embeds Foundry API secrets in customer-facing browser code. Evidence: Exposed credentials could be reused for unauthorized inference. The business goal is to protect model endpoints with least privilege and managed credentials. The proposed solution must respect documented authority and organizational controls. Which approach most directly meets the stated business objective within the approved authority?
- Choose Analyst with reviewed files and independently validate results
- Move secrets server-side and use approved identity and network restrictions
- Extract text using suitable Vision tooling and validate key amounts
- Shortlist catalog options meeting the workload and policy
Explanation
Correct Answer
B:
Exposed credentials could be reused for unauthorized inference. Move secrets server-side and use approved identity and network restrictions.
Incorrect Answers
A:
“Choose Analyst with reviewed files and independently validate results” tests a proposed change for the condition: The task requires calculations and statistical checks beyond simple prose. This question instead requires an action that can protect model endpoints with least privilege and managed credentials. Exposed credentials could be reused for unauthorized inference. Move secrets server-side and use approved identity and network restrictions.
C:
“Extract text using suitable Vision tooling and validate key amounts” is a design or implementation response to the condition: The original files are scans and require text extraction. This question instead requires an action that can protect model endpoints with least privilege and managed credentials. Exposed credentials could be reused for unauthorized inference. Move secrets server-side and use approved identity and network restrictions.
D:
“Shortlist catalog options meeting the workload and policy” is a design or implementation response to the condition: Candidate models differ in modality, region, cost and accuracy. This question instead requires an action that can protect model endpoints with least privilege and managed credentials. Exposed credentials could be reused for unauthorized inference. Move secrets server-side and use approved identity and network restrictions.
Question 708
At Riverbend Publishing, 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. 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 workspace access and unreviewed deployments
- Monitor silent data cleaning errors and misaligned periods
- Monitor index synchronization and overbroad search access
- Monitor changed access lists and index filter bypass
Explanation
Correct Answer
C:
A model's static training data cannot reliably track those files. Monitor index synchronization and overbroad search access. 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:
“Monitor workspace access and unreviewed deployments” monitors this signal: The team must compare models, tools and safety controls as one lifecycle. 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. Monitor index synchronization and overbroad search access. 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:
“Monitor silent data cleaning errors and misaligned periods” monitors this signal: The task requires calculations and statistical checks beyond simple prose. 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. Monitor index synchronization and overbroad search access. 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.
D:
“Monitor changed access lists and index filter bypass” monitors this signal: Index query results lack caller-specific authorization filters. 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. Monitor index synchronization and overbroad search access. 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 709
At Juniper Operations, a department boasts that every employee finished an AI training video. Evidence: Video completion does not show safe real-world ability. The business goal is to measure actual capability and workflow outcomes. The proposed solution must respect documented authority and organizational controls. Which approach most directly meets the stated business objective within the approved authority?
- Use practical assessments with feedback and refresher learning
- Assign eligible plans or consumption options based on approved demand
- Compare pay-as-you-go with available commitments over realistic scenarios
- Use only approved datasets with documented rights and minimization
Explanation
Correct Answer
A:
Video completion does not show safe real-world ability. Use practical assessments with feedback and refresher learning.
Incorrect Answers
B:
“Assign eligible plans or consumption options based on approved demand” is a design or implementation response to the condition: Usage patterns vary and some workflows may use included chat capabilities. This question instead requires an action that can measure actual capability and workflow outcomes. Video completion does not show safe real-world ability. Use practical assessments with feedback and refresher learning.
C:
“Compare pay-as-you-go with available commitments over realistic scenarios” is a design or implementation response to the condition: Demand is uncertain and production volume may be seasonal. This question instead requires an action that can measure actual capability and workflow outcomes. Video completion does not show safe real-world ability. Use practical assessments with feedback and refresher learning.
D:
“Use only approved datasets with documented rights and minimization” is a design or implementation response to the condition: Rights to read documents do not automatically permit model training. This question instead requires an action that can measure actual capability and workflow outcomes. Video completion does not show safe real-world ability. Use practical assessments with feedback and refresher learning.
Question 710
At Atlas Design, a team tests its AI assistant using only polite expected queries. Evidence: Malicious or unexpected inputs could bypass intended restrictions. The business goal is to include adversarial testing proportionate to risk. 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?
- Collect concerns and map the tasks that still require expertise
- Identify licenses, retention requirements and authorized purposes
- List likely prompt injection, impersonation and tool misuse scenarios
- Classify severity and identify source, recipients and active actions
Explanation
Correct Answer
C:
Malicious or unexpected inputs could bypass intended restrictions. List likely prompt injection, impersonation and tool misuse scenarios.
Incorrect Answers
A:
“Collect concerns and map the tasks that still require expertise” is a diagnostic step for the reported condition: Leadership communication focuses only on headcount cost savings. This question instead requires an action that can include adversarial testing proportionate to risk. Malicious or unexpected inputs could bypass intended restrictions. List likely prompt injection, impersonation and tool misuse scenarios.
B:
“Identify licenses, retention requirements and authorized purposes” is a diagnostic step for the reported condition: Rights to read documents do not automatically permit model training. This question instead requires an action that can include adversarial testing proportionate to risk. Malicious or unexpected inputs could bypass intended restrictions. List likely prompt injection, impersonation and tool misuse scenarios.
D:
“Classify severity and identify source, recipients and active actions” is a diagnostic step for the reported condition: There is no owner or process for AI-specific incidents. This question instead requires an action that can include adversarial testing proportionate to risk. Malicious or unexpected inputs could bypass intended restrictions. List likely prompt injection, impersonation and tool misuse scenarios.