Question 150
A design adds autonomous behavior to a process where two customers store similarly named support incidents. This creates a documented risk: Search might mix tenant records because isolation relies only on prompt wording. Which enforced control is the best fit?
- Never rely on prompt text alone for tenancy isolation
- Revoke shared credentials and forbid cross-stage access
- Retain audit records of source replacements and withdrawals
- Keep enough prior approved artifacts for recovery
Explanation
Correct Answer
A:The business setting is two customers store similarly named support incidents, with the exposure that search might mix tenant records because isolation relies only on prompt wording; the enforceable protection must correspond to the path where partitioning and authorized filters constrain relevant candidates.
Incorrect Answers
B:This restriction is associated with test automation must not reach live customer records and addresses the exposure that staging credentials might allow writes to production business systems; it does not specifically protect the target workflow against search might mix tenant records because isolation relies only on prompt wording.
C:Another process could need this guardrail because a disputed answer might be untraceable to its source and index revision; the design being assessed supports two customers store similarly named support incidents and has the explicit exposure that search might mix tenant records because isolation relies only on prompt wording.
D:A preprocessing release changes category labels incorrectly; its guardrail addresses the possibility that a rollback might restore a model but not its required preprocessing artifacts; the present architecture must constrain the different threat that search might mix tenant records because isolation relies only on prompt wording.
Question 151
A solution architecture is being reviewed for this business situation: Prompt wording persuades the agent to submit a privileged operation. The observed defect is that API trusts the model's text assertion of permission. Which architectural decision addresses the underlying requirement?
- Constrain high-risk tools with action-level policy checks
- Separate read-only grounding from transactional tools
- Test prompt changes on paraphrases and edge conditions
- Require independent backend authorization for generated tool calls
Explanation
Correct Answer
D:The defect is that API trusts the model's text assertion of permission; the expected business case is that prompt wording persuades the agent to submit a privileged operation; the design fits because server-side access checks enforce real actor entitlements.
Incorrect Answers
A:There is a valid place for the mechanism that execution-time destination and permission checks block exfiltration, namely a prompt injection requests data export to an outside address; it does not remove the specific defect that API trusts the model's text assertion of permission from this case.
B:The technical behavior of this alternative is that distinct permission surfaces enforce least privilege by task; the user case being assessed is prompt wording persuades the agent to submit a privileged operation with a different failure: API trusts the model's text assertion of permission.
C:This design would make sense if a rewritten prompt improves a single demonstration, particularly where slightly different wording breaks correct entity selection; the requested architecture instead needs the dependency that server-side access checks enforce real actor entitlements because API trusts the model's text assertion of permission.
Question 152
A planned AI solution must support this situation: The task follows fixed approval branches with no interpretation needed. Its prototype fails because employees receive inconsistent approvals when prose drives the decision. Which design best fits the constraint?
- Use an autonomous agent only for bounded reversible tasks
- Constrain agent tools by explicit allowlists and permissions
- Use a deterministic Power Automate flow for stable approvals
- Separate retrieved data from trusted system instructions
Explanation
Correct Answer
C:The architecture must reconcile two facts: the task follows fixed approval branches with no interpretation needed, yet employees receive inconsistent approvals when prose drives the decision; the selected pattern does so through the mechanism that explicit conditions execute known business rules predictably and leave an auditable branch history.
Incorrect Answers
A:For work can proceed from events without human initiation but needs bounded autonomy, this other approach relies on the fact that event triggers and tool permissions can automate predefined safe steps within explicit limits; the issue here is employees receive inconsistent approvals when prose drives the decision; that mechanism does not establish the required explicit conditions execute known business rules predictably and leave an auditable branch history.
B:For another type of architecture, allowlisting and server-enforced identity restrict actual executable capability is useful; this design must instead support the task follows fixed approval branches with no interpretation needed despite the problem where employees receive inconsistent approvals when prose drives the decision.
D:For another type of architecture, trust boundaries treat source text as evidence rather than authority is useful; this design must instead support the task follows fixed approval branches with no interpretation needed despite the problem where employees receive inconsistent approvals when prose drives the decision.
Question 153
An enterprise team is making a platform design decision. Business context: General models repeatedly misread rare engineering labels. Present failure: Prompt changes fail to resolve a measured specialist error. Which approach resolves the mismatch?
- Invoke an AI-backed action from a canvas app with typed inputs
- Version models and datasets together for controlled adaptation
- Use a declarative agent for role-tailored Microsoft 365 work
- Select a custom model for evidence-backed specialized extraction
- Customize service triage around business-specific severity criteria
Explanation
Correct Answer
D:The mechanism that domain-trained models can be evaluated when generic baselines fall short fits the scenario where general models repeatedly misread rare engineering labels; it addresses the defect that prompt changes fail to resolve a measured specialist error.
Incorrect Answers
A:For inspectors submit a short description and need suggested categorization, this other approach relies on the fact that an app action validates inputs and structured results before saving; the issue here is prompt changes fail to resolve a measured specialist error; that mechanism does not establish the required domain-trained models can be evaluated when generic baselines fall short.
B:When retraining uses revised labels and a different source snapshot, dataset lineage and model versioning support repeatable assessment; in the stated workflow, however, prompt changes fail to resolve a measured specialist error; the relevant architectural behavior is that domain-trained models can be evaluated when generic baselines fall short.
C:The technical behavior of this alternative is that declarative agents define instructions, knowledge and actions in the supported host; the user case being assessed is general models repeatedly misread rare engineering labels with a different failure: prompt changes fail to resolve a measured specialist error.
E:When a specialized support team uses a nonstandard severity matrix, custom business rules align AI-assisted triage with actual impact definitions; in the stated workflow, however, prompt changes fail to resolve a measured specialist error; the relevant architectural behavior is that domain-trained models can be evaluated when generic baselines fall short.
Question 154
Implementation alternatives must be compared for a workflow where a third-party AI tool changes where submitted documents are processed. The pilot shows this gap: Approved data crosses an unreviewed processor. Which alternative should lead the solution?
- Audit privileged agent actions with correlation to user intent
- Verify downstream vendor terms and data handling before integration
- Map all agent data flows to approved geographic boundaries
- Regression-test case routing after service agent changes
Explanation
Correct Answer
B:Without the intended behavior, approved data crosses an unreviewed processor; the proposed architecture establishes the specific design principle that contract and privacy assessment cover service location and subprocessor use for the process where a third-party AI tool changes where submitted documents are processed.
Incorrect Answers
A:Logs show deletion but not the requesting principal or approval; the alternate design can address that situation because end-to-end audit binds user, tool, affected record and authorization event, whereas this question needs a solution to approved data crosses an unreviewed processor.
C:For another type of architecture, data-flow architecture identifies processing, storage and telemetry locations is useful; this design must instead support a third-party AI tool changes where submitted documents are processed despite the problem where approved data crosses an unreviewed processor.
D:The technical behavior of this alternative is that controlled service evaluations validate routing before rollout; the user case being assessed is a third-party AI tool changes where submitted documents are processed with a different failure: approved data crosses an unreviewed processor.
Question 155
Support engineers report this failure: Answers adopt an instruction embedded in a source document. The business context is: Third-party documents include sentences telling the model to ignore rules. What evidence should be collected to isolate the cause?
- Document specific requirements unmet by available platforms
- Compare source update timestamps with ingestion checkpoints
- Inspect traces for instruction-following from retrieved content
- Correlate retrieval timeouts with unsupported responses
Explanation
Correct Answer
C:Where answers adopt an instruction embedded in a source document, the useful evidence comes from inspecting traces for instruction-following from retrieved content; its focus matches the architectural mechanism that trust boundaries treat source text as evidence rather than authority in the process where third-party documents include sentences telling the model to ignore rules.
Incorrect Answers
A:For the failure that a boxed product cannot express the execution and ownership rules, documenting specific requirements unmet by available platforms is relevant; this question instead reports that answers adopt an instruction embedded in a source document, so different evidence is needed.
B:The evidence this method produces applies to a workflow where operational records update faster than the agent's indexed copy; it cannot by itself establish why answers adopt an instruction embedded in a source document in the current scenario.
D:In the separate scenario where required knowledge stores are intermittently unreachable, the same investigation could explain the agent fabricates an answer during a search outage; it does not isolate the reported cause in a system where third-party documents include sentences telling the model to ignore rules.
Question 156
An enterprise team is making a platform design decision. Business context: A newly deployed Foundry agent has no detailed diagnostics. Present failure: The on-call team cannot explain failed tool calls. Which approach resolves the mismatch?
- Integrate runtime tracing and alerting into release readiness
- Regression-test case routing after service agent changes
- Version agent instructions, tool definitions and model references together
- Preserve customer communication approvals across solution releases
Explanation
Correct Answer
A:The defect is that the on-call team cannot explain failed tool calls; the expected business case is that a newly deployed Foundry agent has no detailed diagnostics; the design fits because connected observability supplies traces and operational metrics.
Incorrect Answers
B:The technical behavior of this alternative is that controlled service evaluations validate routing before rollout; the user case being assessed is a newly deployed Foundry agent has no detailed diagnostics with a different failure: the on-call team cannot explain failed tool calls.
C:For model and tool changes are promoted independently, this other approach relies on the fact that a complete agent release manifest pins related runtime artifacts; the issue here is the on-call team cannot explain failed tool calls; that mechanism does not establish the required connected observability supplies traces and operational metrics.
D:The alternative addresses new prompt automation bypasses supervisor approval by a design in which business approval checks remain independent of agent version; it does not supply the missing relationship for a newly deployed Foundry agent has no detailed diagnostics, where the on-call team cannot explain failed tool calls.
Question 157
The operating team must monitor whether an architectural correction works. Business context: General models repeatedly misread rare engineering labels. Historical symptom: Prompt changes fail to resolve a measured specialist error. Which outcome measure is the best match?
- Accepted structured suggestions with correct record linkage
- Successful inferences within latency SLO
- Improvement on domain-specific holdout accuracy
- Grounding recall for critical enterprise questions
Explanation
Correct Answer
C:For the business use where general models repeatedly misread rare engineering labels, the result to quantify is improvement on domain-specific holdout accuracy; it connects measurement to the mechanism where domain-trained models can be evaluated when generic baselines fall short, not merely to agent traffic.
Incorrect Answers
A:For the situation where inspectors submit a short description and need suggested categorization, accepted structured suggestions with correct record linkage is a sensible indicator; the problem that prompt changes fail to resolve a measured specialist error requires measurement of improvement on domain-specific holdout accuracy.
B:For the situation where a specialized model must serve production apps reliably, successful inferences within latency SLO is a sensible indicator; the problem that prompt changes fail to resolve a measured specialist error requires measurement of improvement on domain-specific holdout accuracy.
D:The alternative tracks grounding recall for critical enterprise questions, arising from the mechanism where labeled queries and expected evidence distinguish retrieval from synthesis; its movement alone says little about the failure that prompt changes fail to resolve a measured specialist error.
Question 158
Risk reviewers need a control, not merely a monitoring dashboard. The use case is: New behavior must be evaluated without exposing the full workforce. Potential harm: A new agent release could regress behavior across the entire employee base. Which protection fits?
- Set enforceable monetary and scope thresholds outside prompts
- Keep a known-good model and agent configuration for rollback
- Set usage budgets, cost alerts and routing limits
- Scope connector permissions to specific business operations
Explanation
Correct Answer
B:When a new agent release could regress behavior across the entire employee base, the use case new behavior must be evaluated without exposing the full workforce requires a guardrail aligned with the behavior that controlled release gates restrict blast radius and support rollback.
Incorrect Answers
A:The alternate boundary mitigates a model could override a monetary limit through conversational instructions, which is relevant to the agent may refund large amounts or deactivate accounts; it leaves the current workflow vulnerable to a new agent release could regress behavior across the entire employee base.
C:That protection is tied to the mechanism that workload forecasts include model usage, connectors, storage and observability; it does not prevent the distinct risk where a new agent release could regress behavior across the entire employee base in the system under review.
D: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 a new agent release could regress behavior across the entire employee base remains relevant to the present assessment.
Question 159
The operating team must monitor whether an architectural correction works. Business context: Agents must distinguish an incident, complaint, inquiry and entitlement. Historical symptom: Copilot misclassifies a complaint as an ordinary information request. Which outcome measure is the best match?
- Fresh authorized account insights from external systems
- Sensitive-field disclosure rate in connector responses
- Unauthorized financial commitment language rate
- Correct identification of governed service intents
Explanation
Correct Answer
D:The observed shortcoming is that Copilot misclassifies a complaint as an ordinary information request; measuring correct identification of governed service intents gives business owners a directly relevant indicator for the process where agents must distinguish an incident, complaint, inquiry and entitlement.
Incorrect Answers
A:For the situation where sellers need current credit exposure held outside CRM, fresh authorized account insights from external systems is a sensible indicator; the problem that Copilot misclassifies a complaint as an ordinary information request requires measurement of correct identification of governed service intents.
B:A change in sensitive-field disclosure rate in connector responses is informative for sales copilots should not expose sensitive margin assumptions; it does not establish whether the current process achieves correct identification of governed service intents.
C:Although unauthorized financial commitment language rate is observable, the affected process involves agents must distinguish an incident, complaint, inquiry and entitlement; its acceptance outcome is correct identification of governed service intents, not the alternative measurement.
Question 160
A release must prove that pure semantic search misses precise part identifiers will not recur. The application is used in this situation: Product catalog queries mix synonyms and exact serial codes. Which acceptance test most directly checks that failure mode?
- Try off-script wording and verify the mandatory sequence
- Simulate failed verification and confirm no notification
- Query both model codes and paraphrased product descriptions
- Test scans with tables, handwriting and missing labels
Explanation
Correct Answer
C:Querying both model codes and paraphrased product descriptions is aligned with the failure that pure semantic search misses precise part identifiers; its causal relevance is that hybrid ranking combines lexical precision and semantic relevance for the situation where product catalog queries mix synonyms and exact serial codes.
Incorrect Answers
A:Although the exercise might catch generative routing skips a mandatory disclosure step, it would not directly validate the mechanism that hybrid ranking combines lexical precision and semantic relevance in the target implementation.
B:Although the exercise might catch an unconstrained planner sends a notification before authorization, it would not directly validate the mechanism that hybrid ranking combines lexical precision and semantic relevance in the target implementation.
D:For vendors send variable-layout invoices and shipping documents, the alternate test is about whether document processing models extract structured fields from document content; it cannot prove the outcome needed after the reported regression where pure semantic search misses precise part identifiers.