Question 1180
A risk analysis highlights a threat in this workflow: An employee requests a standard equipment return initiation. Exposure: A single-purpose return agent might invoke unrelated inventory or purchasing tools. Which enforceable safeguard addresses it at the relevant boundary?
- Limit tool availability and confirmations for risky steps
- Require least-privilege service permissions and action confirmation
- Give the agent only the required return operation
- Keep business-critical formulas out of free-form generation
Explanation
Correct Answer
C:In the context where an employee requests a standard equipment return initiation, the exposure is that a single-purpose return agent might invoke unrelated inventory or purchasing tools; containing it requires a guardrail in the operation where single-purpose tools and typed parameters bound the task.
Incorrect Answers
A:The alternate guard addresses a generative planner might chain tools that exceed the approved workflow scope for users request a status check and an authorized follow-up action; the present threat, a single-purpose return agent might invoke unrelated inventory or purchasing tools, needs a different enforcement boundary.
B:The alternate guard addresses a tool might update tickets using a shared identity with too many permissions for an agent needs to update a governed ticket status; the present threat, a single-purpose return agent might invoke unrelated inventory or purchasing tools, needs a different enforcement boundary.
D:The alternate guard addresses free-form model arithmetic might change a critical service-level calculation for a page must show a reproducible service-level metric; the present threat, a single-purpose return agent might invoke unrelated inventory or purchasing tools, needs a different enforcement boundary.
Question 1181
A planned AI solution must support this situation: A user asks the agent to read an unrelated executive record. Its prototype fails because the model may try a tool outside the user's rights. Which design best fits the constraint?
- Constrain high-risk tools with action-level policy checks
- Test tool-permission escalation attempts at the execution layer
- Test concurrent agent runs for duplicate side effects
- Separate read-only grounding from transactional tools
Explanation
Correct Answer
B:The architecture must reconcile two facts: a user asks the agent to read an unrelated executive record, yet the model may try a tool outside the user's rights; the selected pattern does so through the mechanism that authorization evaluations verify server-side deny behavior.
Incorrect Answers
A:In the case where a prompt injection requests data export to an outside address, the alternate design depends on execution-time destination and permission checks block exfiltration; it does not correct the target failure, the model may try a tool outside the user's rights.
C:For another type of architecture, concurrency evaluation tests idempotency and serialization is useful; this design must instead support a user asks the agent to read an unrelated executive record despite the problem where the model may try a tool outside the user's rights.
D:This design would make sense if an agent should summarize account data but not alter it, particularly where retrieval configuration accidentally grants write capability; the requested architecture instead needs the dependency that authorization evaluations verify server-side deny behavior because the model may try a tool outside the user's rights.
Question 1182
A product owner must measure whether an AI-enabled change actually improves this process: A voice bot cannot resolve a complex eligibility dispute. The earlier problem was: Customers repeat authentication and issue details after transfer. Which outcome indicator is most relevant?
- Transfers resolved without repeated customer explanation
- Unauthorized ancillary actions per task
- Channel-appropriate responses passing usability review
- Complex exception tasks resolved with supported reasoning traces
Explanation
Correct Answer
A:Transfers resolved without repeated customer explanation; this outcome measure belongs to the workflow where a voice bot cannot resolve a complex eligibility dispute; a change in it can help detect whether the defect that customers repeat authentication and issue details after transfer is resolved.
Incorrect Answers
B:For the workflow where a bot only needs to check warranty and prepare a draft, unauthorized ancillary actions per task is an actionable measure; it does not establish whether customers repeat authentication and issue details after transfer in the situation under review.
C:A change in channel-appropriate responses passing usability review is informative for the same agent is offered in voice and asynchronous chat; it does not establish whether the current process achieves transfers resolved without repeated customer explanation.
D:For the workflow where procurement exceptions require comparing several policies and records, complex exception tasks resolved with supported reasoning traces is an actionable measure; it does not establish whether customers repeat authentication and issue details after transfer in the situation under review.
Question 1183
An agent-assisted service does not behave as specified: Retrieval mixes records from two organizations. It is used in this situation: Many tenants have identically named projects. Which observable diagnostic procedure is best?
- Review action arguments and created escalation ownership
- Compare extraction failures by document layout
- Trace statements to referenced financial data
- Examine query filters and retrieved tenant IDs
Explanation
Correct Answer
D:The reported fault is that retrieval mixes records from two organizations in a system where many tenants have identically named projects; evidence from examining query filters and retrieved tenant IDs can show whether tenant and entity filters constrain candidate records before generation.
Incorrect Answers
A:The evidence this method produces applies to a workflow where a representative must initiate an exception workflow from the case; it cannot by itself establish why retrieval mixes records from two organizations in the current scenario.
B:For the failure that template parsing omits line items on unfamiliar forms, comparing extraction failures by document layout is relevant; this question instead reports that retrieval mixes records from two organizations, so different evidence is needed.
C:The proposed probe concerns finance asks why a variance was flagged, where grounded explanatory responses distinguish evidence from hypothesis; this incident instead requires evidence about tenant and entity filters constrain candidate records before generation and the failure that retrieval mixes records from two organizations.
Question 1184
An enterprise team is making a platform design decision. Business context: Users must remain in a proprietary work-order application. Present failure: Requiring a separate chat portal increases context switching. Which approach resolves the mismatch?
- Buy a supported managed capability for standard requirements
- Use Microsoft 365 Copilot connectors for external knowledge
- Embed a Copilot-powered interaction inside an existing line-of-business app
- Use MCP for an agent's access to external tools and data
Explanation
Correct Answer
C:Users must remain in a proprietary work-order application; its failing behavior is that requiring a separate chat portal increases context switching; applying the underlying design principle—app-hosted integration brings relevant AI actions into an owned user interface—connects the requirement to an implementable control.
Incorrect Answers
A:Managed functionality shifts routine platform operations to the service; its main purpose is the requirement is common and product coverage is adequate, not correction of the reported defect where requiring a separate chat portal increases context switching in the target process.
B:This design would make sense if teams want approved external records searchable in Copilot, particularly where the agent lacks knowledge of non-Microsoft content; the requested architecture instead needs the dependency that app-hosted integration brings relevant AI actions into an owned user interface because requiring a separate chat portal increases context switching.
D:In the case where a Copilot Studio agent needs a reusable interface to governed operations, the alternate design depends on MCP presents supported tools and resources through a common integration contract; it does not correct the target failure, requiring a separate chat portal increases context switching.
Question 1185
The operating team must monitor whether an architectural correction works. Business context: Field staff must keep working during a model outage. Historical symptom: Submission screens freeze when inference fails. Which outcome measure is the best match?
- Accepted structured suggestions with correct record linkage
- Core form completion rate during AI service incidents
- Compound requests completed with correct tool sequence
- Case-workspace completion without navigation detours
Explanation
Correct Answer
B:The use case involves field staff must keep working during a model outage and the baseline failure that submission screens freeze when inference fails; the KPI tracks core form completion rate during AI service incidents, so improvement has an interpretable operational meaning.
Incorrect Answers
A:The alternative tracks accepted structured suggestions with correct record linkage, arising from the mechanism where an app action validates inputs and structured results before saving; its movement alone says little about the failure that submission screens freeze when inference fails.
C:The alternative tracks compound requests completed with correct tool sequence, arising from the mechanism where the planner can select supported knowledge and tools for compound intents; its movement alone says little about the failure that submission screens freeze when inference fails.
D:The proposed metric concerns staff need a contextual workspace assembled around one customer case, whereas the failure in this case is that submission screens freeze when inference fails; the indicator needed here is core form completion rate during AI service incidents.
Question 1186
During integration testing, engineers tune prompts without checking whether evidence was retrieved. The associated business context is: An answer contradicts the handbook despite high model fluency. Which architectural remedy is the strongest match?
- Tune descriptions and tool constraints before adding more tools
- Use grounded answer evaluation to separate retrieval and generation faults
- Use identity-aware caching for grounded agent responses
- Inspect retrieval spans when cited answers omit evidence
Explanation
Correct Answer
B:The architecture must reconcile two facts: an answer contradicts the handbook despite high model fluency, yet engineers tune prompts without checking whether evidence was retrieved; the selected pattern does so through the mechanism that separate relevance and groundedness tests isolate the failing layer.
Incorrect Answers
A:This design would make sense if planner calls the wrong tool among overlapping capabilities, particularly where the team proposes adding another similar tool; the requested architecture instead needs the dependency that separate relevance and groundedness tests isolate the failing layer because engineers tune prompts without checking whether evidence was retrieved.
C:There is a valid place for the mechanism that cache keys and access validation prevent security-context leakage, namely shared result caching makes answers fast across roles; it does not remove the specific defect that engineers tune prompts without checking whether evidence was retrieved from this case.
D:The alternative addresses surface quality scores appear healthy despite wrong documents by a design in which retrieval telemetry distinguishes evidence search from answer generation; it does not supply the missing relationship for an answer contradicts the handbook despite high model fluency, where engineers tune prompts without checking whether evidence was retrieved.
Question 1187
A planned AI solution must support this situation: A prompt injection requests data export to an outside address. Its prototype fails because model-level refusal alone may not prevent a tool call. Which design best fits the constraint?
- Constrain high-risk tools with action-level policy checks
- Treat retrieved documents as untrusted evidence rather than instructions
- Inspect model-call timing separately from integration latency
- Tune descriptions and tool constraints before adding more tools
Explanation
Correct Answer
A:The architecture must reconcile two facts: a prompt injection requests data export to an outside address, yet model-level refusal alone may not prevent a tool call; the selected pattern does so through the mechanism that execution-time destination and permission checks block exfiltration.
Incorrect Answers
B:For another type of architecture, trust hierarchy separates lower-trust content from agent instructions is useful; this design must instead support a prompt injection requests data export to an outside address despite the problem where model-level refusal alone may not prevent a tool call.
C:In the case where agents regularly exceed an interactive response-time objective, the alternate design depends on per-span latency isolates model work, network waits and tool execution; it does not correct the target failure, model-level refusal alone may not prevent a tool call.
D:Clear capabilities and restricted sets improve selection reliability; its main purpose is planner calls the wrong tool among overlapping capabilities, not correction of the reported defect where model-level refusal alone may not prevent a tool call in the target process.
Question 1188
Security and business owners want to keep this capability: Sales and service agents need the same customer and case context. They must also prevent the following: Cross-application aggregation might expose records outside the user's Dynamics permissions. Which guardrail should the design incorporate?
- Scope connector permissions to specific business operations
- Keep a known-good model and agent configuration for rollback
- Require review for breaking changes and publish migration windows
- Honor cross-app role permissions and data minimization
- Prevent duplicate side effects during task retries
Explanation
Correct Answer
D:In the scenario where sales and service agents need the same customer and case context, the explicit risk is that cross-application aggregation might expose records outside the user's Dynamics permissions; enforcement must follow the mechanism by which shared authorized records support coordinated business context.
Incorrect Answers
A:Where users need familiar Copilot behavior plus an internal status lookup, the guard can address an extension could receive broad write permissions beyond its single new action; the exposure that cross-application aggregation might expose records outside the user's Dynamics permissions remains relevant to the present assessment.
B:The safeguard for new behavior must be evaluated without exposing the full workforce deals with a different threat: a new agent release could regress behavior across the entire employee base; even after it is enforced, the exposure that cross-application aggregation might expose records outside the user's Dynamics permissions remains possible in the target workflow where shared authorized records support coordinated business context.
C:For the use case where multiple AI teams consume a changing source API, the control addresses the threat that an API schema change could break consuming agents without migration notice; the specified agent must be protected against cross-application aggregation might expose records outside the user's Dynamics permissions.
E:The safeguard for downstream research agents run longer than the interactive request deals with a different threat: retried delegated work could charge a customer or create a record more than once; even after it is enforced, the exposure that cross-application aggregation might expose records outside the user's Dynamics permissions remains possible in the target workflow where shared authorized records support coordinated business context.
Question 1189
One enterprise rollout has 2 separately testable requirements. (1) Business use: Deployment automation pulls model artifacts from unverified origins. Known defect: A replaced model package introduces unknown behavior. (2) Business use: Proprietary records are copied to an open test workspace. Known defect: Training examples become accessible outside intended teams. Which 2 regression tests directly satisfy these stated cases? Choose TWO.
- Replace a candidate hash in staging and verify rejection
- Try accessing a restricted training corpus from an unrelated project
- Query identical terms as HR and non-HR users
- Simulate rollback during a nonproduction close cycle
Explanation
Correct Answers
A:For the business workflow where deployment automation pulls model artifacts from unverified origins, replacing a candidate hash in staging and verifying rejection challenges the failure that a replaced model package introduces unknown behavior; the release also needs a separate check for proprietary records are copied to an open test workspace.
B:For the business workflow where proprietary records are copied to an open test workspace, trying accessing a restricted training corpus from an unrelated project challenges the failure that training examples become accessible outside intended teams; the release also needs a separate check for deployment automation pulls model artifacts from unverified origins.
Incorrect Answers
C:This regression test is associated with the different failure where agent cites a private HR document to unauthorized staff in a workflow where users have different rights to the same knowledge corpus; it does not directly solve the listed conditions: a replaced model package introduces unknown behavior; training examples become accessible outside intended teams.
D:This regression test is associated with the different failure where a prompt change during close generates unexplained exceptions in a workflow where invoice posting agents affect critical close processes; it does not directly solve the listed conditions: a replaced model package introduces unknown behavior; training examples become accessible outside intended teams.
Question 1190
The fault affects an AI-powered workflow in which an intake task requires several prerequisite fields. Specifically, agent sends incomplete requests when users answer out of order. Which diagnostic investigation best distinguishes the relevant failure path?
- Compare displayed metrics with independently computed results
- Compare outputs against approved language baselines
- Compare cross-service commit histories and compensation events
- Inspect missing-field collection and final action payload
Explanation
Correct Answer
D:The reported fault is that agent sends incomplete requests when users answer out of order in a system where an intake task requires several prerequisite fields; evidence from inspecting missing-field collection and final action payload can show whether state tracking captures required inputs before any action.
Incorrect Answers
A:In the process where a page must show a reproducible service-level metric, the diagnostic procedure could explain lLM-generated arithmetic changes across otherwise identical prompts; this particular incident instead turns on whether state tracking captures required inputs before any action and why agent sends incomplete requests when users answer out of order.
B:The proposed inspection is associated with compliance letters must preserve approved language and disclaimers, where templating and approved clauses limit generated variation; it would not expose the failure mode that agent sends incomplete requests when users answer out of order with the required specificity.
C:In the process where an agent creates an order but reservation of inventory fails, the diagnostic procedure could explain half-complete transactions leave operational state inconsistent; this particular incident instead turns on whether state tracking captures required inputs before any action and why agent sends incomplete requests when users answer out of order.