Question 1160
One enterprise rollout has 2 separately testable requirements. (1) Business use: Personal read-only assistants and payment agents have different risk. Exposure: A privileged autonomous agent might receive the same review as a private helper. (2) Business use: Business units create agents with no central visibility. Exposure: Production agents might have no recorded accountable owner or risk classification. Which 2 safeguards directly satisfy these stated cases? Choose TWO.
- Require heightened review for autonomous privileged agents
- Restore prior version on any critical schema or safety failure
- Block publication without accountable owner and classification
- Require reviewed datasets and independent promotion sign-off
Explanation
Correct Answers
A:For the exposure that a privileged autonomous agent might receive the same review as a private helper, zoned governance aligns controls with impact and data sensitivity explains the protection required when personal read-only assistants and payment agents have different risk; this does not replace control of production agents might have no recorded accountable owner or risk classification.
C:For the exposure that production agents might have no recorded accountable owner or risk classification, inventory metadata records owner, risk tier, tools and lifecycle state explains the protection required when business units create agents with no central visibility; this does not replace control of a privileged autonomous agent might receive the same review as a private helper.
Incorrect Answers
B:This security safeguard addresses the unrelated exposure that a canary model might cause critical schema or safety regressions without quick recovery in a workflow where business applications depend on the agent endpoint continuously; the question instead names the specific risks a privileged autonomous agent might receive the same review as a private helper; production agents might have no recorded accountable owner or risk classification; this choice does not directly mitigate those specific exposures.
D:This security safeguard addresses the unrelated exposure that unreviewed training examples might poison a promoted model in a workflow where business analysts can submit examples but not change production behavior; the question instead names the specific risks a privileged autonomous agent might receive the same review as a private helper; production agents might have no recorded accountable owner or risk classification; this choice does not directly mitigate those specific exposures.
Question 1161
The architect wants evidence before altering the deployment. In a workflow where users use novel requests outside the supported topic inventory, the agent repeats an unhelpful default message. What should the team examine first?
- Count fallback utterances and escalation outcomes
- Review action inputs, examples and schema-parse logs
- Validate indexed records and source ACL mappings
- Review agent capability manifest against requested functions
- Inspect step run histories and conditional branch outcomes
Explanation
Correct Answer
A:The reported fault is that the agent repeats an unhelpful default message in a system where users use novel requests outside the supported topic inventory; evidence from counting fallback utterances and escalation outcomes can show whether fallback handles unmatched intent and can ask for clarification or escalate.
Incorrect Answers
B:Reviewing action inputs, examples and schema-parse logs explores a separate connection, where prompt actions encapsulate reusable inputs and expected outputs; the diagnosed requirement is users use novel requests outside the supported topic inventory, and the actual defect is the agent repeats an unhelpful default message.
C:The evidence this method produces applies to a workflow where workers need internal wiki articles from outside Microsoft 365; it cannot by itself establish why the agent repeats an unhelpful default message in the current scenario.
D:The proposed inspection is associated with employees want curated assistance without owning a separate LLM runtime, where declarative agents define instructions, knowledge and actions in the supported host; it would not expose the failure mode that the agent repeats an unhelpful default message with the required specificity.
E:That probe evaluates defined flow steps make sequencing and error paths explicit in a workflow where approval processing requires sequential verification and notification; the reported failure concerns the path for which fallback handles unmatched intent and can ask for clarification or escalate instead.
Question 1162
A product owner must measure whether an AI-enabled change actually improves this process: The orchestrator chooses between similar invoice and credit tools. The earlier problem was: It passes customer IDs into supplier inputs. Which outcome indicator is most relevant?
- Correct tool selection and input-binding rate
- Correctly sequenced workflow executions
- Unauthorized data fields sent to model actions
- Validated entity and sentiment tagging accuracy
Explanation
Correct Answer
A:For the business use where the orchestrator chooses between similar invoice and credit tools, the result to quantify is correct tool selection and input-binding rate; it connects measurement to the mechanism where specific tool metadata helps the planner choose and parameterize actions, not merely to agent traffic.
Incorrect Answers
B:Although correctly sequenced workflow executions is observable, the affected process involves the orchestrator chooses between similar invoice and credit tools; its acceptance outcome is correct tool selection and input-binding rate, not the alternative measurement.
C:Although unauthorized data fields sent to model actions is observable, the affected process involves the orchestrator chooses between similar invoice and credit tools; its acceptance outcome is correct tool selection and input-binding rate, not the alternative measurement.
D:The premise of this indicator is purpose-built text analysis handles bounded language-analysis tasks; the architecture being evaluated instead serves the orchestrator chooses between similar invoice and credit tools, and its success is measured by correct tool selection and input-binding rate.
Question 1163
A business stakeholder describes this operating situation: Customers believe generated advice is a human decision. Testing instead shows that users cannot identify automated recommendations. What design approach is most appropriate?
- Plan reversible rollout for supply-chain decision assistance
- Provide clear disclosure when users interact with AI
- Design meaningful human oversight for consequential recommendations
- Version training datasets with lineage and access records
Explanation
Correct Answer
B:For the need that customers believe generated advice is a human decision, the relevant mechanism is that transparency cues distinguish AI suggestions and human approvals; that architectural dependency addresses the observed failure where users cannot identify automated recommendations.
Incorrect Answers
A:For another type of architecture, a pilot cohort and rollback path isolate operational impact is useful; this design must instead support customers believe generated advice is a human decision despite the problem where users cannot identify automated recommendations.
C:In the case where AI proposes eligibility decisions affecting customers, the alternate design depends on explainable review interfaces support accountable judgment; it does not correct the target failure, users cannot identify automated recommendations.
D:When model retraining consumes periodically revised business records, immutable snapshots and lineage tie runs to approved sources; in the stated workflow, however, users cannot identify automated recommendations; the relevant architectural behavior is that transparency cues distinguish AI suggestions and human approvals.
Question 1164
An incident review has one confirmed symptom: It substitutes a prior customer's order in a later turn. For the use case where agent must remember order context while asking for missing details, which targeted diagnostic check would reveal the most pertinent evidence?
- Reconcile audit IDs across each application
- Compare operation keys with final business transaction counts
- Compare retrieval and tool responses using two user roles
- Inspect conversation state and parameter mapping across turns
Explanation
Correct Answer
D:Where it substitutes a prior customer's order in a later turn, the useful evidence comes from inspecting conversation state and parameter mapping across turns; its focus matches the architectural mechanism that multi-turn tests assess state tracking, entity separation and follow-up questions in the process where agent must remember order context while asking for missing details.
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 agent must remember order context while asking for missing details.
B:For the failure that parallel agents each submit a costly replacement, comparing operation keys with final business transaction counts is relevant; this question instead reports that it substitutes a prior customer's order in a later turn, so different evidence is needed.
C:The evidence this method produces applies to a workflow where sales and Finance roles have different record entitlements; it cannot by itself establish why it substitutes a prior customer's order in a later turn in the current scenario.
Question 1165
A policy review covers an agent for which service agents draft customer messages but cannot send without review. Its most relevant exposure is that a service drafting assistant might gain permission to send unapproved notices. What safeguard should be required before release?
- Separate draft and send permissions in service integrations
- Restrict promotion to reviewed compatible configurations
- Enforce approval before production import and publication
- Block release when production variables or connections are unresolved
Explanation
Correct Answer
A:The business setting is service agents draft customer messages but cannot send without review, with the exposure that a service drafting assistant might gain permission to send unapproved notices; the enforceable protection must correspond to the path where business approval checks remain independent of agent version.
Incorrect Answers
B:Model and tool changes are promoted independently; its guardrail addresses the possibility that an agent release might pair instructions with incompatible tool or model versions; the present architecture must constrain the different threat that a service drafting assistant might gain permission to send unapproved notices.
C:For the use case where the same agent must move from development to test and production, the control addresses the threat that a maker might import or publish an agent without release approval; the specified agent must be protected against a service drafting assistant might gain permission to send unapproved notices.
D:The alternate guard addresses imported actions might retain staging endpoints after production deployment for a flow moved to production still points to a test endpoint; the present threat, a service drafting assistant might gain permission to send unapproved notices, needs a different enforcement boundary.
Question 1166
An enterprise team is making a platform design decision. Business context: Policy answers must be tied to a current source. Present failure: The agent confidently answers when retrieval fails. Which approach resolves the mismatch?
- Constrain high-risk tools with action-level policy checks
- Test grounded prompts for citation and abstention behavior
- Check bias and performance across represented populations
- Test prompt changes on paraphrases and edge conditions
Explanation
Correct Answer
B:Because evidence and refusal evaluations enforce source-supported claims, this approach corrects the reported breakdown where the agent confidently answers when retrieval fails; the affected situation is that policy answers must be tied to a current source.
Incorrect Answers
A:The alternative addresses model-level refusal alone may not prevent a tool call by a design in which execution-time destination and permission checks block exfiltration; it does not supply the missing relationship for policy answers must be tied to a current source, where the agent confidently answers when retrieval fails.
C:In the case where a classifier underperforms on one language group, the alternate design depends on segmented evaluations reveal uneven model behavior; it does not correct the target failure, the agent confidently answers when retrieval fails.
D: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 evidence and refusal evaluations enforce source-supported claims because the agent confidently answers when retrieval fails.
Question 1167
A staging test has to detect the specific defect that the same agent can read an invoice and commit a payment unchecked. In the workflow where teams propose that an agent perform both low-risk lookups and financial postings, which exercise verifies the corrective behavior?
- Update a case and verify an authorized sales workflow sees it
- Remove required sources and check that the agent declines
- Run a pilot with a comparable holdout group
- Attempt low-risk and high-risk actions and inspect approval enforcement
Explanation
Correct Answer
D:The check is targeted rather than generic: tiered permissions require progressively stronger approvals for riskier actions; its pass/fail evidence applies to the workflow where teams propose that an agent perform both low-risk lookups and financial postings and to the prior defect that the same agent can read an invoice and commit a payment unchecked.
Incorrect Answers
A:The verification concerning sales and service agents need the same customer and case context checks shared authorized records support coordinated business context against an opportunity owner sees a stale service escalation status; its result says little about the observed issue where the same agent can read an invoice and commit a payment unchecked.
B:A pass on that verification would speak to the system cannot always find evidence for regulated guidance; it would not show that tiered permissions require progressively stronger approvals for riskier actions, which is needed to avoid the observed defect the same agent can read an invoice and commit a payment unchecked.
C:In a solution where management attributes a sales lift to Copilot without a comparison, running a pilot with a comparable holdout group could catch the defect that seasonality produces the same change in unassisted teams; the current release gate must challenge the failure where the same agent can read an invoice and commit a payment unchecked.
Question 1168
An investment review asks what measurable result would justify continuing an AI solution where employees install external agents with unknown data pathways. The original issue is: An unreviewed plugin exports sensitive tenant information. Which indicator should be tracked?
- Third-party processing paths with verified compliance records
- Third-party agents completing security review before activation
- Retired agents with no active unintended access
- Model releases containing unapproved training records
Explanation
Correct Answer
B:Third-party agents completing security review before activation; this outcome measure belongs to the workflow where employees install external agents with unknown data pathways; a change in it can help detect whether the defect that an unreviewed plugin exports sensitive tenant information is resolved.
Incorrect Answers
A:For the situation where a third-party AI tool changes where submitted documents are processed, third-party processing paths with verified compliance records is a sensible indicator; the problem that an unreviewed plugin exports sensitive tenant information requires measurement of third-party agents completing security review before activation.
C:The premise of this indicator is decommissioning revokes credentials, connectors and publishing rights; the architecture being evaluated instead serves employees install external agents with unknown data pathways, and its success is measured by third-party agents completing security review before activation.
D:That scorecard concerns the activity in which business analysts can submit examples but not change production behavior and its result of model releases containing unapproved training records; the reported shortcoming here is an unreviewed plugin exports sensitive tenant information.
Question 1169
A policy review covers an agent for which the planner calls an order-cancel tool for status requests. Its most relevant exposure is that an erroneous planner might invoke an unapproved privileged tool. What safeguard should be required before release?
- Stop unapproved tools from executing regardless of selected plan
- Use bounded retries and approved fallback capacity
- Redact identifiers and restrict transcript access
- Redact customer data before making examples available to developers
Explanation
Correct Answer
A:The business setting is the planner calls an order-cancel tool for status requests, with the exposure that an erroneous planner might invoke an unapproved privileged tool; the enforceable protection must correspond to the path where step traces expose planning, selected tools and arguments.
Incorrect Answers
B:The proposed control targets retry storms might amplify a rate-limit event into a service outage, while the given risk scenario is an erroneous planner might invoke an unapproved privileged tool; these exposures arise in different operational contexts.
C:Where debugging logs include prompts containing customer data, the guard can address agent transcripts might expose direct identifiers to roles without a legitimate need; the exposure that an erroneous planner might invoke an unapproved privileged tool remains relevant to the present assessment.
D:The proposed control targets user transcripts pasted into developer tickets might disclose customer details, while the given risk scenario is an erroneous planner might invoke an unapproved privileged tool; these exposures arise in different operational contexts.
Question 1170
The pilot needs a durable correction: New planning behavior occasionally leaks internal notes. This affects the workflow in which a hosted agent gains an additional research tool. Which design choice should the architect approve?
- Preserve customer communication approvals across solution releases
- Monitor newly released service agent behavior by customer cohort
- Integrate runtime tracing and alerting into release readiness
- Use preproduction evaluation gates for agent workflow changes
- Test posting permissions and approvals after agent updates
Explanation
Correct Answer
D:For the need that a hosted agent gains an additional research tool, the relevant mechanism is that evaluation suites validate quality and safety before publication; that architectural dependency addresses the observed failure where new planning behavior occasionally leaks internal notes.
Incorrect Answers
A:In the case where service agents draft customer messages but cannot send without review, the alternate design depends on business approval checks remain independent of agent version; it does not correct the target failure, new planning behavior occasionally leaks internal notes.
B:Cohort telemetry supports safe staged validation; its main purpose is a conversational change affects only one support geography, not correction of the reported defect where new planning behavior occasionally leaks internal notes in the target process.
C:In the case where a newly deployed Foundry agent has no detailed diagnostics, the alternate design depends on connected observability supplies traces and operational metrics; it does not correct the target failure, new planning behavior occasionally leaks internal notes.
E:For agents suggest payment adjustments and inventory holds, this other approach relies on the fact that authorization regression gates verify posting authority remains constrained; the issue here is new planning behavior occasionally leaks internal notes; that mechanism does not establish the required evaluation suites validate quality and safety before publication.