Question 1080
A policy review covers an agent for which users have different rights to the same knowledge corpus. Its most relevant exposure is that shared answer caches might drop source ACLs and leak a private document. What safeguard should be required before release?
- Require heightened review for autonomous privileged agents
- Retain audit records of source replacements and withdrawals
- Reject cached results that omit original security context
- Do not allow model confidence prose to override missing evidence
- Block publishing when required evidence cannot be recovered
Explanation
Correct Answer
C:Avoiding the consequence where shared answer caches might drop source ACLs and leak a private document requires a control for users have different rights to the same knowledge corpus; in this architecture, retrieval applies effective source entitlements to candidate documents, which is the point at which the boundary can be applied.
Incorrect Answers
A:Another process could need this guardrail because a privileged autonomous agent might receive the same review as a private helper; the design being assessed supports users have different rights to the same knowledge corpus and has the explicit exposure that shared answer caches might drop source ACLs and leak a private document.
B:This restriction is associated with agents begin citing an obsolete article after an ingestion job and addresses the exposure that a disputed answer might be untraceable to its source and index revision; it does not specifically protect the target workflow against shared answer caches might drop source ACLs and leak a private document.
D:The alternate guard addresses confident model language might override an explicit evidence requirement for policy answers must be tied to a current source; the present threat, shared answer caches might drop source ACLs and leak a private document, needs a different enforcement boundary.
E:An answer contradicts the handbook despite high model fluency; its guardrail addresses the possibility that the team might publish a model that never retrieves the evidence it requires; the present architecture must constrain the different threat that shared answer caches might drop source ACLs and leak a private document.
Question 1081
Before introducing a fix, the team must explain why full replacement increases retraining and migration risk. The affected setting is: Current users need one governed custom action inside their workflow. What should be inspected?
- Compare required outcomes with the prebuilt agent's supported boundaries
- Review available tool inventory versus approved actions
- Compare baseline model errors by language and domain
- Compare gap size to supported extension interfaces
Explanation
Correct Answer
D:For the situation where current users need one governed custom action inside their workflow, the defect is that full replacement increases retraining and migration risk; comparing gap size to supported extension interfaces helps verify the cause-and-effect path: extension retains familiar user experience and adds bounded custom behavior.
Incorrect Answers
A:In the process where users need ordinary meeting summaries and follow-up drafts, the diagnostic procedure could explain a costly custom runtime replicates a mature standard capability; this particular incident instead turns on whether extension retains familiar user experience and adds bounded custom behavior and why full replacement increases retraining and migration risk.
B:Agent instructions mention many APIs but only two are permitted; its matching diagnostic check is reasonable for the agent can invoke an unintended high-privilege operation; the present investigation needs evidence for the different failure where full replacement increases retraining and migration risk.
C:The proposed inspection is associated with a unique domain vocabulary produces persistent measured errors, where a measured capability gap, appropriate dataset and MLOps plan justify customization; it would not expose the failure mode that full replacement increases retraining and migration risk with the required specificity.
Question 1082
An agent rollout is intended for this situation: Support leads need resolution and escalation insights over time. The known problem is that team measures chat volume but cannot see failed conversations. Which success measure connects to that specific outcome?
- Complete distributed traces per sampled transaction
- Unauthorized production customization events
- Resolved session rate alongside escalation and abandonment
- Preproduction regressions caught before agent publication
Explanation
Correct Answer
C:The observed shortcoming is that team measures chat volume but cannot see failed conversations; measuring resolved session rate alongside escalation and abandonment gives business owners a directly relevant indicator for the process where support leads need resolution and escalation insights over time.
Incorrect Answers
A:Complete distributed traces per sampled transaction; that indicator describes a distinct workflow where a request crosses Copilot Studio, a flow and the CRM backend; it does not quantify resolved session rate alongside escalation and abandonment, which is needed for the stated case.
B:For the workflow where production components should not be edited casually, unauthorized production customization events is an actionable measure; it does not establish whether team measures chat volume but cannot see failed conversations in the situation under review.
D:Although preproduction regressions caught before agent publication is observable, the affected process involves support leads need resolution and escalation insights over time; its acceptance outcome is resolved session rate alongside escalation and abandonment, not the alternative measurement.
Question 1083
An enterprise team is making a platform design decision. Business context: An employee requests a standard equipment return initiation. Present failure: A broad assistant changes unrelated inventory records. Which approach resolves the mismatch?
- Choose bounded task-specific instructions over open-ended autonomy
- Build a scoped task agent for repeatable knowledge-to-action work
- Use generative orchestration for multi-intent tool composition
- Provide source-linked explanations for analytical responses
Explanation
Correct Answer
B:Because single-purpose tools and typed parameters bound the task, this approach corrects the reported breakdown where a broad assistant changes unrelated inventory records; the affected situation is that an employee requests a standard equipment return initiation.
Incorrect Answers
A:For another type of architecture, limited intent and tool scope prevent unnecessary actions is useful; this design must instead support an employee requests a standard equipment return initiation despite the problem where a broad assistant changes unrelated inventory records.
C:When users request a status check and an authorized follow-up action, the planner can select supported knowledge and tools for compound intents; in the stated workflow, however, a broad assistant changes unrelated inventory records; the relevant architectural behavior is that single-purpose tools and typed parameters bound the task.
D:When finance asks why a variance was flagged, grounded explanatory responses distinguish evidence from hypothesis; in the stated workflow, however, a broad assistant changes unrelated inventory records; the relevant architectural behavior is that single-purpose tools and typed parameters bound the task.
Question 1084
Before introducing a fix, the team must explain why a client can request arbitrary portions of training examples. The affected setting is: An internal model sometimes returns confidential text verbatim. What should be inspected?
- Audit training-data writers and model approvers
- Review backend authorization logs against actor identities
- Trace source exception through target state and error queue
- Review blocked-output annotations and access attempts
Explanation
Correct Answer
D:A client can request arbitrary portions of training examples; to attribute the defect rather than guess at a fix, reviewing blocked-output annotations and access attempts examines the dependency where safety controls and authorization checks limit misuse and disclosure.
Incorrect Answers
A:Auditing training-data writers and model approvers explores a separate connection, where separated data-curation and release permissions preserve integrity; the diagnosed requirement is an internal model sometimes returns confidential text verbatim, and the actual defect is a client can request arbitrary portions of training examples.
B:The proposed inspection is associated with prompt wording persuades the agent to submit a privileged operation, where server-side access checks enforce real actor entitlements; it would not expose the failure mode that a client can request arbitrary portions of training examples with the required specificity.
C:In the process where a blocked invoice triggers a hold on related procurement actions, the diagnostic procedure could explain a network fault leaves the two apps inconsistent; this particular incident instead turns on whether safety controls and authorization checks limit misuse and disclosure and why a client can request arbitrary portions of training examples.
Question 1085
An investment review asks what measurable result would justify continuing an AI solution where agents return inconsistent fields for downstream integrations. The original issue is: Narrative answers cannot be parsed by consuming workflows. Which indicator should be tracked?
- Authorized structured-record answer accuracy
- Format compliance improvement on unseen tasks
- Correct classification rate for ambiguous boundary cases
- Schema-valid response rate
Explanation
Correct Answer
D:The use case involves agents return inconsistent fields for downstream integrations and the baseline failure that narrative answers cannot be parsed by consuming workflows; the KPI tracks schema-valid response rate, so improvement has an interpretable operational meaning.
Incorrect Answers
A:The proposed metric concerns users ask about governed CRM records rather than prose manuals, whereas the failure in this case is that narrative answers cannot be parsed by consuming workflows; the indicator needed here is schema-valid response rate.
B:A change in format compliance improvement on unseen tasks is informative for a stable specialized classification style persists despite prompt examples; it does not establish whether the current process achieves schema-valid response rate.
C:Although correct classification rate for ambiguous boundary cases is observable, the affected process involves agents return inconsistent fields for downstream integrations; its acceptance outcome is schema-valid response rate, not the alternative measurement.
Question 1086
A proposed deployment supports this business case: A retrieved email contains instructions to bypass approval. Reviewers also warn that injected text might persuade a model to invoke a protected API. Which restriction mitigates that concrete risk?
- Block protected actions unless the trusted user and backend authorize
- Prevent actions before identity and missing fields are verified
- Fail closed on authorization and schema-validation failures
- Reject missing, invalid or out-of-scope authorization tokens
Explanation
Correct Answer
A:When injected text might persuade a model to invoke a protected API, the use case a retrieved email contains instructions to bypass approval requires a guardrail aligned with the behavior that adversarial fixtures verify separation of sources and instructions.
Incorrect Answers
B:The safeguard for agent must remember order context while asking for missing details deals with a different threat: an agent might act before collecting mandatory identity and form fields; even after it is enforced, the exposure that injected text might persuade a model to invoke a protected API remains possible in the target workflow where adversarial fixtures verify separation of sources and instructions.
C:For the use case where benchmarks cover only cooperative user wording, the control addresses the threat that malformed calls might pass a happy-path suite yet mutate records; the specified agent must be protected against injected text might persuade a model to invoke a protected API.
D:Prompt wording persuades the agent to submit a privileged operation; its guardrail addresses the possibility that an API might trust model-generated identity claims instead of valid authorization; the present architecture must constrain the different threat that injected text might persuade a model to invoke a protected API.
Question 1087
A policy review covers an agent for which retrieved policy fragments lose context when documents are split. Its most relevant exposure is that chunked content might lose original permissions and version information. What safeguard should be required before release?
- Remove unneeded attributes before model invocation
- Stop and compensate when a required approval fails
- Confirm identity before sharing protected case details
- Retain document version and ACL information on every chunk
Explanation
Correct Answer
D:Retrieved policy fragments lose context when documents are split; its risk scenario is that chunked content might lose original permissions and version information; the safeguard fits because semantically coherent chunks with provenance preserve context during retrieval and the restriction addresses that specific exposure.
Incorrect Answers
A:The safeguard for a canvas app passes customer records to a model action deals with a different threat: a canvas app might send unnecessary protected fields into an inference prompt; even after it is enforced, the exposure that chunked content might lose original permissions and version information remains possible in the target workflow where semantically coherent chunks with provenance preserve context during retrieval.
B:The alternate guard addresses a flow might notify the customer even though a required approval failed for approval processing requires sequential verification and notification; the present threat, chunked content might lose original permissions and version information, needs a different enforcement boundary.
C:That protection is tied to the mechanism that structured handoff passes allowed intent, transcript summary and case context; it does not prevent the distinct risk where chunked content might lose original permissions and version information in the system under review.
Question 1088
A pilot scorecard needs evidence of business improvement rather than only invocation counts. The use case is: Production documents differ from clean training samples. Known defect: Scanned forms suddenly produce inaccurate predictions. Which KPI should be selected?
- Detected attacks blocked before unsafe execution
- Worst-performing segment error rate
- Model accuracy under agreed input-quality stress levels
- Unauthorized action rate in adversarial evaluations
Explanation
Correct Answer
C:This scorecard needs an outcome related to production documents differ from clean training samples; model accuracy under agreed input-quality stress levels. That indicator can be checked against the earlier symptom that scanned forms suddenly produce inaccurate predictions.
Incorrect Answers
A:That scorecard concerns the activity in which hostile instructions arrive both as user input and retrieved documents and its result of detected attacks blocked before unsafe execution; the reported shortcoming here is scanned forms suddenly produce inaccurate predictions.
B:The alternative tracks worst-performing segment error rate, arising from the mechanism where segmented evaluations reveal uneven model behavior; its movement alone says little about the failure that scanned forms suddenly produce inaccurate predictions.
D:For the workflow where benchmarks cover only cooperative user wording, unauthorized action rate in adversarial evaluations is an actionable measure; it does not establish whether scanned forms suddenly produce inaccurate predictions in the situation under review.
Question 1089
A planned AI solution must support this situation: Agent can research options but cannot approve a vendor payment. Its prototype fails because a background process posts expenses beyond its delegated scope. Which design best fits the constraint?
- Use connector actions for authenticated data mutations
- Trigger an autonomous agent from a governed business event
- Use a prompt action for bounded text transformation
- Set autonomy boundaries with approval checkpoints
- Use agent reasoning for uncertain multi-step analysis with bounded tools
Explanation
Correct Answer
D:The architecture must reconcile two facts: agent can research options but cannot approve a vendor payment, yet a background process posts expenses beyond its delegated scope; the selected pattern does so through the mechanism that approval checkpoints split independent analysis from high-impact execution.
Incorrect Answers
A:There is a valid place for the mechanism that an authorized connector action invokes an operational service, namely an agent needs to update a governed ticket status; it does not remove the specific defect that a background process posts expenses beyond its delegated scope from this case.
B:There is a valid place for the mechanism that an event trigger starts the bounded agent workflow on state change, namely shipment exceptions should initiate investigation without a chat message; it does not remove the specific defect that a background process posts expenses beyond its delegated scope from this case.
C:The alternative addresses a general conversational turn produces inconsistent output fields by a design in which prompt actions encapsulate reusable inputs and expected outputs; it does not supply the missing relationship for agent can research options but cannot approve a vendor payment, where a background process posts expenses beyond its delegated scope.
E:This design would make sense if procurement exceptions require comparing several policies and records, particularly where a simple scripted topic cannot handle legitimate variants; the requested architecture instead needs the dependency that approval checkpoints split independent analysis from high-impact execution because a background process posts expenses beyond its delegated scope.
Question 1090
Security and business owners want to keep this capability: Answers depend on policies revised every week. They must also prevent the following: A model might state outdated rules as facts when authoritative policies have changed. Which guardrail should the design incorporate?
- Disallow factual claims unsupported by current policy data
- Use approved extension points rather than unsupported modifications
- Promote complex or low-confidence cases to stronger fallback
- Add human approval for consequential recommendations
Explanation
Correct Answer
A:When a model might state outdated rules as facts when authoritative policies have changed, the use case answers depend on policies revised every week requires a guardrail aligned with the behavior that retrieval separates volatile knowledge from model parameters.
Incorrect Answers
B:Another process could need this guardrail because an unsupported product customization might break after vendor updates; the design being assessed supports answers depend on policies revised every week and has the explicit exposure that a model might state outdated rules as facts when authoritative policies have changed.
C:The safeguard for many calls require only a small fixed-label output deals with a different threat: a small model might make a high-complexity judgment with insufficient confidence; even after it is enforced, the exposure that a model might state outdated rules as facts when authoritative policies have changed remains possible in the target workflow where retrieval separates volatile knowledge from model parameters.
D:Another process could need this guardrail because an automated recommendation might be executed without human accountability; the design being assessed supports answers depend on policies revised every week and has the explicit exposure that a model might state outdated rules as facts when authoritative policies have changed.