Question 810
A quality engineer is writing a regression case after the incident where universal use of a premium model inflates cost. The underlying business scenario is: Many calls require only a small fixed-label output. What should the test exercise?
- Compare matched assisted and unassisted case cohorts
- Demonstrate a custom multi-step workflow and failure recovery
- Prototype a unique scheduling dependency with the custom runtime
- Benchmark a simple-task holdout on both model tiers
- Compare prioritized roadmap with a risk-adjusted benefit matrix
Explanation
Correct Answer
D:Benchmarking a simple-task holdout on both model tiers is aligned with the failure that universal use of a premium model inflates cost; its causal relevance is that task-aware routing chooses economical models after quality gates for the situation where many calls require only a small fixed-label output.
Incorrect Answers
A:Comparing matched assisted and unassisted case cohorts verifies an adjacent objective where AI drafts requests quickly but specialists must correct some outputs; it leaves untested the causal path in which task-aware routing chooses economical models after quality gates for this case.
B:In the alternate setting where the solution demands its own model selection and hosting policies, the test exercises custom-engine architecture controls runtime, model calls and integrations; even a pass would not show that universal use of a premium model inflates cost has been prevented.
C:The suggested test serves a workflow needs proprietary scheduling, model choice and runtime control, in which a predefined host cannot meet the required orchestration contract; the present scenario instead concerns the reported problem where universal use of a premium model inflates cost.
E:The suggested test serves candidate processes are attractive but permissions and data quality are weak, in which the first-wave pilot requests broad access to unreliable records; the present scenario instead concerns the reported problem where universal use of a premium model inflates cost.
Question 811
An operational risk could arise because a canary model might cause critical schema or safety regressions without quick recovery. The AI system supports the business need that business applications depend on the agent endpoint continuously. Which guardrail is designed for this threat?
- Restore prior version on any critical schema or safety failure
- Rotate compromised keys and enforce least-privilege callers
- Prevent training with unapproved or expired sensitive data
- Block publication without accountable owner and classification
Explanation
Correct Answer
A:Avoiding the consequence where a canary model might cause critical schema or safety regressions without quick recovery requires a control for business applications depend on the agent endpoint continuously; in this architecture, canary traffic and versioned rollback reduce release risk, which is the point at which the boundary can be applied.
Incorrect Answers
B:The alternate guard addresses a leaked endpoint credential might permit inference from an untrusted client for a sensitive model endpoint is reachable using a broadly shared key; the present threat, a canary model might cause critical schema or safety regressions without quick recovery, needs a different enforcement boundary.
C:Another process could need this guardrail because training could ingest data whose use rights expired or were never approved; the design being assessed supports business applications depend on the agent endpoint continuously and has the explicit exposure that a canary model might cause critical schema or safety regressions without quick recovery.
D:The safeguard for business units create agents with no central visibility deals with a different threat: production agents might have no recorded accountable owner or risk classification; even after it is enforced, the exposure that a canary model might cause critical schema or safety regressions without quick recovery remains possible in the target workflow where canary traffic and versioned rollback reduce release risk.
Question 812
One enterprise rollout has 2 separately testable requirements. (1) Business use: An AI triage model under-serves one customer language. Known defect: Overall performance masks unequal escalation rates. (2) Business use: A helpful assistant can still produce unsafe financial or legal guidance. Known defect: Pilot satisfaction hides severe low-frequency failures. Which 2 diagnostic procedures directly satisfy these stated cases? Choose TWO.
- Review critical failure classes and mitigation owners
- Review ranking against incident impact and affected populations
- Compare error and escalation metrics across justified cohorts
- Reconcile tenant deployments with the approved registry
Explanation
Correct Answers
A:Reviewing critical failure classes and mitigation owners can reveal why pilot satisfaction hides severe low-frequency failures when a helpful assistant can still produce unsafe financial or legal guidance; that finding is needed alongside the independent diagnosis for an AI triage model under-serves one customer language.
C:Comparing error and escalation metrics across justified cohorts can reveal why overall performance masks unequal escalation rates when an AI triage model under-serves one customer language; that finding is needed alongside the independent diagnosis for a helpful assistant can still produce unsafe financial or legal guidance.
Incorrect Answers
B:This diagnosis is associated with the different failure where a rare data leak risk remains below cosmetic response issues in a workflow where product owners favor easy fixes over high-consequence failures; it does not directly solve the listed conditions: overall performance masks unequal escalation rates; pilot satisfaction hides severe low-frequency failures.
D:This diagnosis is associated with the different failure where security cannot identify which agents use protected connectors in a workflow where business units create agents with no central visibility; it does not directly solve the listed conditions: overall performance masks unequal escalation rates; pilot satisfaction hides severe low-frequency failures.
Question 813
A quality engineer is writing a regression case after the incident where transcription confusion risks wrong account modifications. The underlying business scenario is: Callers provide account numbers and amounts by speech. What should the test exercise?
- Drive a recommendation above the threshold and confirm blocking
- Change a permitted credit field and verify updated authorized retrieval
- Simulate similar-sounding identifiers and validate confirmation
- Install the agent to a pilot Microsoft 365 audience
Explanation
Correct Answer
C:Simulating similar-sounding identifiers and validating confirmation is aligned with the failure that transcription confusion risks wrong account modifications; its causal relevance is that spoken confirmation and verification reduce high-impact recognition errors for the situation where callers provide account numbers and amounts by speech.
Incorrect Answers
A:Although the exercise might catch a background process posts expenses beyond its delegated scope, it would not directly validate the mechanism that spoken confirmation and verification reduce high-impact recognition errors in the target implementation.
B:This check exercises a different system behavior: a supported connector can retrieve external information with explicit identity mapping; a pass there cannot establish that transcription confusion risks wrong account modifications no longer occurs in the required workflow.
D:For employees want curated assistance without owning a separate LLM runtime, the alternate test is about whether declarative agents define instructions, knowledge and actions in the supported host; it cannot prove the outcome needed after the reported regression where transcription confusion risks wrong account modifications.
Question 814
A policy review covers an agent for which a Teams user should query and update a supported external service. Its most relevant exposure is that a plugin might create business records without user consent or approval. What safeguard should be required before release?
- Limit creation privileges and prevent duplicate escalations
- Require explicit consent for consequential create or update operations
- Prevent writes or disclosure outside user permissions
- Assign glossary approval to service process owners
Explanation
Correct Answer
B:A Teams user should query and update a supported external service; its risk scenario is that a plugin might create business records without user consent or approval; the safeguard fits because a plugin provides authorized callable tools to an agent and the restriction addresses that specific exposure.
Incorrect Answers
A:The alternate guard addresses repeated agent calls might create duplicate escalation records or excess privileges for a representative must initiate an exception workflow from the case; the present threat, a plugin might create business records without user consent or approval, needs a different enforcement boundary.
C:Finance users need current operational data in contextual help; its guardrail addresses the possibility that live finance records might become editable or visible to unauthorized user roles; the present architecture must constrain the different threat that a plugin might create business records without user consent or approval.
D:The proposed control targets unreviewed synonyms could redefine legally significant customer-service classifications, while the given risk scenario is a plugin might create business records without user consent or approval; these exposures arise in different operational contexts.
Question 815
A risk analysis highlights a threat in this workflow: Users need ordinary meeting summaries and follow-up drafts. Exposure: A standard agent might violate tenant access policy when activated for many users. Which enforceable safeguard addresses it at the relevant boundary?
- Review tenancy, data terms, exit and support commitments
- Keep proprietary calculations and secrets on the service side
- Narrow allowed tool capabilities and verify server-side authorization
- Confirm tenant policy and data access before enabling adoption
Explanation
Correct Answer
D:Avoiding the consequence where a standard agent might violate tenant access policy when activated for many users requires a control for users need ordinary meeting summaries and follow-up drafts; in this architecture, a prebuilt Microsoft 365 capability minimizes custom maintenance for commodity tasks, which is the point at which the boundary can be applied.
Incorrect Answers
A:The proposed control targets a managed service might fail a required data-processing or exit condition, while the given risk scenario is a standard agent might violate tenant access policy when activated for many users; these exposures arise in different operational contexts.
B:For the use case where standard conversation needs a proprietary scoring service, the control addresses the threat that a third-party conversational host might receive proprietary scoring logic or secrets; the specified agent must be protected against a standard agent might violate tenant access policy when activated for many users.
C:Where a Copilot Studio agent needs a reusable interface to governed operations, the guard can address a shared MCP server might honor an unauthorized tool invocation; the exposure that a standard agent might violate tenant access policy when activated for many users remains relevant to the present assessment.
Question 816
A solution architecture is being reviewed for this business situation: Network retries can repeat a single purchase request. The observed defect is that two purchase orders appear from one user intent. Which architectural decision addresses the underlying requirement?
- Make task actions retry-safe through idempotency
- Define robust tool descriptions and input schemas for planner selection
- Use MCP tools for a governed reusable integration surface
- Build a scoped task agent for repeatable knowledge-to-action work
Explanation
Correct Answer
A:Without the intended behavior, two purchase orders appear from one user intent; the proposed architecture establishes the specific design principle that idempotency keys tie repeated submissions to one business operation for the process where network retries can repeat a single purchase request.
Incorrect Answers
B:For another type of architecture, specific tool metadata helps the planner choose and parameterize actions is useful; this design must instead support network retries can repeat a single purchase request despite the problem where two purchase orders appear from one user intent.
C:A supported MCP server exposes typed tool capabilities to clients; its main purpose is several agents need the same inventory query operation, not correction of the reported defect where two purchase orders appear from one user intent in the target process.
D:This design would make sense if an employee requests a standard equipment return initiation, particularly where a broad assistant changes unrelated inventory records; the requested architecture instead needs the dependency that idempotency keys tie repeated submissions to one business operation because two purchase orders appear from one user intent.
Question 817
A risk analysis highlights a threat in this workflow: Search spans internal manuals with different owners and permissions. Exposure: Ownerless confidential drafts could become searchable before classification. Which enforceable safeguard addresses it at the relevant boundary?
- Refuse definitive policy advice without a current authoritative citation
- Quarantine unclassified or ownerless content pending review
- Restrict access through source permissions, not prompt wording
- Require idempotency at cross-service mutation boundaries
Explanation
Correct Answer
B:In the scenario where search spans internal manuals with different owners and permissions, the explicit risk is that ownerless confidential drafts could become searchable before classification; enforcement must follow the mechanism by which metadata enables filtering, review, retention and traceable answers.
Incorrect Answers
A:Where policy documents contain draft, superseded, and currently effective revisions, the guard can address draft or superseded policy text might be used as definitive regulated guidance; the exposure that ownerless confidential drafts could become searchable before classification remains relevant to the present assessment.
C:Employee policies reside in governed Microsoft 365 documents; its guardrail addresses the possibility that an agent prompt could claim permission to read a private SharePoint document; the present architecture must constrain the different threat that ownerless confidential drafts could become searchable before classification.
D:For the use case where copilot, Dataverse, Dynamics and custom APIs share one process, the control addresses the threat that network retries could create duplicated writes across independent systems; the specified agent must be protected against ownerless confidential drafts could become searchable before classification.
Question 818
A staging test has to detect the specific defect that repeated fallback messages trap callers in a loop. In the workflow where contact Center agents sometimes cannot resolve complex billing intent, which exercise verifies the corrective behavior?
- Create an escalation from a test case and inspect audit records
- Test mixed intents with missing action parameters
- Trigger unknown requests and verify timely escalation
- Install the agent to a pilot Microsoft 365 audience
Explanation
Correct Answer
C:The historical failure is that repeated fallback messages trap callers in a loop; for the setting where contact Center agents sometimes cannot resolve complex billing intent, triggering unknown requests and verifying timely escalation verifies the dependency where confidence, retry ceilings and escalation routing enable resolution.
Incorrect Answers
A:The suggested test serves a representative must initiate an exception workflow from the case, in which AI suggestions are never converted into traceable escalation records; the present scenario instead concerns the reported problem where repeated fallback messages trap callers in a loop.
B:In a solution where users request a status check and an authorized follow-up action, testing mixed intents with missing action parameters could catch the defect that a single-topic design cannot combine both operations naturally; the current release gate must challenge the failure where repeated fallback messages trap callers in a loop.
D:The verification concerning employees want curated assistance without owning a separate LLM runtime checks declarative agents define instructions, knowledge and actions in the supported host against a custom service would duplicate built-in host orchestration; its result says little about the observed issue where repeated fallback messages trap callers in a loop.
Question 819
An agent-assisted service does not behave as specified: Supervisors routinely click approve without seeing evidence. It is used in this situation: AI proposes eligibility decisions affecting customers. Which observable diagnostic procedure is best?
- Reconcile audit IDs across each application
- Inspect vector-query filters and tenant IDs of returned hits
- Compare ticket assignment confusion matrices across versions
- Compare reviewer decisions with evidence access and override rates
- Compare expected answers against source-of-truth rules
Explanation
Correct Answer
D:The investigation directly tests the causal connection: explainable review interfaces support accountable judgment; it collects evidence for a workflow where AI proposes eligibility decisions affecting customers but the outcome is instead that supervisors routinely click approve without seeing evidence.
Incorrect Answers
A:In the separate scenario where service recommends a price adjustment requiring finance approval, the same investigation could explain the final order lacks traceability to the AI suggestion and approver; it does not isolate the reported cause in a system where AI proposes eligibility decisions affecting customers.
B:That probe evaluates partitioning and authorized filters constrain relevant candidates in a workflow where two customers store similarly named support incidents; the reported failure concerns the path for which explainable review interfaces support accountable judgment instead.
C:The proposed inspection is associated with a support agent changes how it classifies priority intents, where controlled service evaluations validate routing before rollout; it would not expose the failure mode that supervisors routinely click approve without seeing evidence with the required specificity.
E:For the failure that the automated suite approves the same defect it should detect, comparing expected answers against source-of-truth rules is relevant; this question instead reports that supervisors routinely click approve without seeing evidence, so different evidence is needed.
Question 820
A test plan focuses on a previously missed failure: Overall accuracy conceals disproportionate error. It occurred when a classifier underperforms on one language group. Which check directly validates that behavior?
- Evaluate a representative multilingual holdout
- Test equivalent requests in represented languages
- Try injection, invalid inputs and out-of-scope actions
- Disable knowledge access and verify safe no-evidence behavior
Explanation
Correct Answer
A:The scenario is that a classifier underperforms on one language group; the defect to rule out is that overall accuracy conceals disproportionate error; testing via evaluating a representative multilingual holdout challenges the design dependency where segmented evaluations reveal uneven model behavior.
Incorrect Answers
B:The verification concerning an AI triage model under-serves one customer language checks segment-aware evaluations surface fairness risks against overall performance masks unequal escalation rates; its result says little about the observed issue where overall accuracy conceals disproportionate error.
C:In a solution where benchmarks cover only cooperative user wording, trying injection, invalid inputs and out-of-scope actions could catch the defect that a malicious instruction triggers a privileged tool; the current release gate must challenge the failure where overall accuracy conceals disproportionate error.
D:The suggested test serves policy answers must be tied to a current source, in which the agent confidently answers when retrieval fails; the present scenario instead concerns the reported problem where overall accuracy conceals disproportionate error.