Microsoft AB-100 (Agentic AI Business Solutions Architect) Practice Test 71

Question 700

Governance approves the business purpose but requires control of this exposure: A flow might notify the customer even though a required approval failed. The agent will be used when approval processing requires sequential verification and notification. Which boundary is most defensible?

  1. Disable write tools and label generated material for review
  2. Reject unvalidated model output before ticket creation
  3. Ensure the backend rejects repeat writes for the same intent
  4. Stop and compensate when a required approval fails

Explanation

Correct Answer

D:

Because a flow might notify the customer even though a required approval failed, a guardrail must constrain the actual system behavior: defined flow steps make sequencing and error paths explicit; protecting a response alone cannot enforce that boundary.

Incorrect Answers

A:

The safeguard for marketing specialists need tailored descriptions from existing briefings deals with a different threat: a text-only drafting assistant might be given unnecessary record-writing powers; even after it is enforced, the exposure that a flow might notify the customer even though a required approval failed remains possible in the target workflow where defined flow steps make sequencing and error paths explicit.

B:

The alternate boundary mitigates a malformed structured summary might create corrupt service tickets, which is relevant to a summarizer sends output into a ticket ingestion service; it leaves the current workflow vulnerable to a flow might notify the customer even though a required approval failed.

C:

The safeguard for network retries can repeat a single purchase request deals with a different threat: an API retry might create a second purchase for the same user instruction; even after it is enforced, the exposure that a flow might notify the customer even though a required approval failed remains possible in the target workflow where defined flow steps make sequencing and error paths explicit.

Question 701

An agent rollout is intended for this situation: An imported agent can authenticate in dev but not production. The known problem is that users receive permission errors after successful package import. Which success measure connects to that specific outcome?

  1. Recovery time for a failed connector-action deployment
  2. Authorized tool calls succeeding after promotion
  3. Cross-role unauthorized record access rate
  4. Unauthorized write operations executed by read-only agents

Explanation

Correct Answer

B:

This scorecard needs an outcome related to an imported agent can authenticate in dev but not production; authorized tool calls succeeding after promotion. That indicator can be checked against the earlier symptom that users receive permission errors after successful package import.

Incorrect Answers

A:

A new action version changes how cases are updated; its intended measurement is recovery time for a failed connector-action deployment; substituting that result here would not capture whether users receive permission errors after successful package import has been corrected.

C:

That scorecard concerns the activity in which agent looks up records on behalf of employees with different rights and its result of cross-role unauthorized record access rate; the reported shortcoming here is users receive permission errors after successful package import.

D:

Unauthorized write operations executed by read-only agents; that indicator describes a distinct workflow where an agent should summarize account data but not alter it; it does not quantify authorized tool calls succeeding after promotion, which is needed for the stated case.

Question 702

A pilot scorecard needs evidence of business improvement rather than only invocation counts. The use case is: Service agents draft customer messages but cannot send without review. Known defect: New prompt automation bypasses supervisor approval. Which KPI should be selected?

  1. Concessions with full recommendation-to-posting audit chain
  2. Unauthorized production customization events
  3. Outbound regulated messages with required approvals
  4. Release candidates satisfying critical evaluation thresholds
  5. Customer handoffs free from integration mismatches

Explanation

Correct Answer

C:

The observed shortcoming is that new prompt automation bypasses supervisor approval; measuring outbound regulated messages with required approvals gives business owners a directly relevant indicator for the process where service agents draft customer messages but cannot send without review.

Incorrect Answers

A:

Although concessions with full recommendation-to-posting audit chain is observable, the affected process involves service agents draft customer messages but cannot send without review; its acceptance outcome is outbound regulated messages with required approvals, not the alternative measurement.

B:

The alternative tracks unauthorized production customization events, arising from the mechanism where managed solution deployment supports controlled downstream customization boundaries; its movement alone says little about the failure that new prompt automation bypasses supervisor approval.

D:

The alternative tracks release candidates satisfying critical evaluation thresholds, arising from the mechanism where evaluation suites validate quality and safety before publication; its movement alone says little about the failure that new prompt automation bypasses supervisor approval.

E:

A customer handoff depends on shared record and term mappings; its intended measurement is customer handoffs free from integration mismatches; substituting that result here would not capture whether new prompt automation bypasses supervisor approval has been corrected.

Question 703

A planned AI solution must support this situation: Service, sales and finance jointly decide high-value concessions. Its prototype fails because a suggested discount bypasses delegated approval authority. Which design best fits the constraint?

  1. Provide a unified approval handoff for AI-identified account risks
  2. Sequence AI adoption by data and security maturity
  3. Use Dataverse as the governed handoff for customer cases
  4. Use explicit handoff contracts for asynchronous specialist work

Explanation

Correct Answer

A:

Without the intended behavior, a suggested discount bypasses delegated approval authority; the proposed architecture establishes the specific design principle that cross-app workflow routes the recommendation to the accountable approver for the process where service, sales and finance jointly decide high-value concessions.

Incorrect Answers

B:

For another type of architecture, readiness assessment places safe high-value workloads ahead of risky ones is useful; this design must instead support service, sales and finance jointly decide high-value concessions despite the problem where a suggested discount bypasses delegated approval authority.

C:

In the case where sales and service agents need the same customer and case context, the alternate design depends on shared authorized records support coordinated business context; it does not correct the target failure, a suggested discount bypasses delegated approval authority.

D:

This design would make sense if downstream research agents run longer than the interactive request, particularly where orchestrator returns a partial result as though the workflow finished; the requested architecture instead needs the dependency that cross-app workflow routes the recommendation to the accountable approver because a suggested discount bypasses delegated approval authority.

Question 704

A quality engineer is writing a regression case after the incident where free-form paragraphs break required schema parsing. The underlying business scenario is: A summarizer sends output into a ticket ingestion service. What should the test exercise?

  1. Test empty and malformed input and validate response shape
  2. Replay a timeout after committing a purchase
  3. Ask unrelated questions and verify no supplier action is triggered
  4. Submit authorized and unauthorized return requests

Explanation

Correct Answer

A:

Testing empty and malformed input and validating response shape is aligned with the failure that free-form paragraphs break required schema parsing; its causal relevance is that structured outputs define required fields and allowable values for the situation where a summarizer sends output into a ticket ingestion service.

Incorrect Answers

B:

This check exercises a different system behavior: idempotency keys tie repeated submissions to one business operation; a pass there cannot establish that free-form paragraphs break required schema parsing no longer occurs in the required workflow.

C:

The suggested test serves a bot only needs to check warranty and prepare a draft, in which it autonomously contacts a supplier without approval; the present scenario instead concerns the reported problem where free-form paragraphs break required schema parsing.

D:

A pass on that verification would speak to an employee requests a standard equipment return initiation; it would not show that structured outputs define required fields and allowable values, which is needed to avoid the observed defect free-form paragraphs break required schema parsing.

Question 705

Risk reviewers need a control, not merely a monitoring dashboard. The use case is: Teams want approved external records searchable in Copilot. Potential harm: External indexed knowledge might appear to users who lack source access. Which protection fits?

  1. Map source permissions faithfully before broad indexing
  2. Confirm tenant policy and data access before enabling adoption
  3. Validate sending-agent identity and delegated authorization
  4. Scope connector permissions to specific business operations

Explanation

Correct Answer

A:

In the context where teams want approved external records searchable in Copilot, the exposure is that external indexed knowledge might appear to users who lack source access; containing it requires a guardrail in the operation where connectors expose external enterprise content to supported Microsoft search/Copilot experiences.

Incorrect Answers

B:

Another process could need this guardrail because a standard agent might violate tenant access policy when activated for many users; the design being assessed supports teams want approved external records searchable in Copilot and has the explicit exposure that external indexed knowledge might appear to users who lack source access.

C:

The safeguard for independently hosted agents must exchange tasks and results deals with a different threat: an external agent might accept delegated tasks from a spoofed sender; even after it is enforced, the exposure that external indexed knowledge might appear to users who lack source access remains possible in the target workflow where connectors expose external enterprise content to supported Microsoft search/Copilot experiences.

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 external indexed knowledge might appear to users who lack source access remains relevant to the present assessment.

Question 706

Business leaders reject an activity-only metric for this process: Hostile instructions arrive both as user input and retrieved documents. They need to detect the issue where the agent is exposed to repeated jailbreak patterns. Which performance indicator fits?

  1. Detected attacks blocked before unsafe execution
  2. Sensitive output disclosures in model misuse evaluations
  3. Foundry agent releases with reproducible manifests
  4. Successful indirect prompt-injection attacks

Explanation

Correct Answer

A:

This scorecard needs an outcome related to hostile instructions arrive both as user input and retrieved documents; detected attacks blocked before unsafe execution. That indicator can be checked against the earlier symptom that the agent is exposed to repeated jailbreak patterns.

Incorrect Answers

B:

The proposed metric concerns an internal model sometimes returns confidential text verbatim, whereas the failure in this case is that the agent is exposed to repeated jailbreak patterns; the indicator needed here is detected attacks blocked before unsafe execution.

C:

For the workflow where model and tool changes are promoted independently, Foundry agent releases with reproducible manifests is an actionable measure; it does not establish whether the agent is exposed to repeated jailbreak patterns in the situation under review.

D:

Although successful indirect prompt-injection attacks is observable, the affected process involves hostile instructions arrive both as user input and retrieved documents; its acceptance outcome is detected attacks blocked before unsafe execution, not the alternative measurement.

Question 707

One enterprise rollout has 3 separately testable requirements. (1) Business use: A workflow needs proprietary scheduling, model choice and runtime control. Exposure: A bespoke runtime might be shipped without an accountable security or support team. (2) Business use: Users need familiar Copilot behavior plus an internal status lookup. Exposure: An extension could receive broad write permissions beyond its single new action. (3) Business use: Users need ordinary meeting summaries and follow-up drafts. Exposure: A standard agent might violate tenant access policy when activated for many users. Which 3 safeguards directly satisfy these stated cases? Choose THREE.

  1. Confirm tenant policy and data access before enabling adoption
  2. Budget for independent hosting, safety testing and monitoring
  3. Narrow allowed tool capabilities and verify server-side authorization
  4. Scope connector permissions to specific business operations

Explanation

Correct Answers

A:

For the exposure that a standard agent might violate tenant access policy when activated for many users, a prebuilt Microsoft 365 capability minimizes custom maintenance for commodity tasks explains the protection required when users need ordinary meeting summaries and follow-up drafts; this does not replace control of a bespoke runtime might be shipped without an accountable security or support team; an extension could receive broad write permissions beyond its single new action.

B:

For the exposure that a bespoke runtime might be shipped without an accountable security or support team, owning the engine provides control over models, routing and execution explains the protection required when a workflow needs proprietary scheduling, model choice and runtime control; this does not replace control of an extension could receive broad write permissions beyond its single new action; a standard agent might violate tenant access policy when activated for many users.

D:

For the exposure that an extension could receive broad write permissions beyond its single new action, supported extensibility adds an action while retaining the host's orchestration explains the protection required when users need familiar Copilot behavior plus an internal status lookup; this does not replace control of a bespoke runtime might be shipped without an accountable security or support team; a standard agent might violate tenant access policy when activated for many users.

Incorrect Answers

C:

This security safeguard addresses the unrelated exposure that a shared MCP server might honor an unauthorized tool invocation in a workflow where a Copilot Studio agent needs a reusable interface to governed operations; the question instead names the specific risks a bespoke runtime might be shipped without an accountable security or support team; an extension could receive broad write permissions beyond its single new action; a standard agent might violate tenant access policy when activated for many users; this choice does not directly mitigate those specific exposures.

Question 708

A business stakeholder describes this operating situation: A classifier underperforms on one language group. Testing instead shows that overall accuracy conceals disproportionate error. What design approach is most appropriate?

  1. Test structured outputs against explicit format contracts
  2. Validate custom extraction models by field and document class
  3. Check bias and performance across represented populations
  4. Evaluate outcomes for meaningful demographic or language disparities

Explanation

Correct Answer

C:

The architecture must reconcile two facts: a classifier underperforms on one language group, yet overall accuracy conceals disproportionate error; the selected pattern does so through the mechanism that segmented evaluations reveal uneven model behavior.

Incorrect Answers

A:

This design would make sense if a Copilot prompt feeds records into downstream automation, particularly where generated output occasionally omits required keys; the requested architecture instead needs the dependency that segmented evaluations reveal uneven model behavior because overall accuracy conceals disproportionate error.

B:

The technical behavior of this alternative is that stratified field metrics protect high-impact values; the user case being assessed is a classifier underperforms on one language group with a different failure: overall accuracy conceals disproportionate error.

D:

In the case where an AI triage model under-serves one customer language, the alternate design depends on segment-aware evaluations surface fairness risks; it does not correct the target failure, overall accuracy conceals disproportionate error.

Question 709

A pilot scorecard needs evidence of business improvement rather than only invocation counts. The use case is: Agents regularly exceed an interactive response-time objective. Known defect: Engineers blame generation although the backend is slow. Which KPI should be selected?

  1. Unapproved external destinations reached through agent tools
  2. Incorrect tool selections identified before side effects
  3. Failed agent requests with diagnosable trace paths
  4. Tail latency by model, retrieval and tool phase

Explanation

Correct Answer

D:

For the use case where agents regularly exceed an interactive response-time objective, the indicator tail latency by model, retrieval and tool phase is worth baselining against the earlier issue where engineers blame generation although the backend is slow.

Incorrect Answers

A:

A change in unapproved external destinations reached through agent tools is informative for a prompt injection requests data export to an outside address; it does not establish whether the current process achieves tail latency by model, retrieval and tool phase.

B:

For the situation where the planner calls an order-cancel tool for status requests, incorrect tool selections identified before side effects is a sensible indicator; the problem that engineers blame generation although the backend is slow requires measurement of tail latency by model, retrieval and tool phase.

C:

For the workflow where operations need to correlate prompts, model calls and tools in a Foundry workflow, failed agent requests with diagnosable trace paths is an actionable measure; it does not establish whether engineers blame generation although the backend is slow in the situation under review.

Question 710

Implementation alternatives must be compared for a workflow where AI assistance reads operational records through Dataverse. The pilot shows this gap: An update changes a field mapping and grounded answers become stale. Which alternative should lead the solution?

  1. Plan reversible rollout for supply-chain decision assistance
  2. Test Finance-to-Supply-Chain exception handling after partial failure
  3. Pin approved models and dependencies in a reviewed registry
  4. Validate finance and operations virtual-entity mappings after change

Explanation

Correct Answer

D:

AI assistance reads operational records through Dataverse; its failing behavior is that an update changes a field mapping and grounded answers become stale; applying the underlying design principle—integration tests protect virtual-entity schema and source semantics—connects the requirement to an implementable control.

Incorrect Answers

A:

The alternative addresses recommendation changes spread to all planners without evaluation by a design in which a pilot cohort and rollback path isolate operational impact; it does not supply the missing relationship for AI assistance reads operational records through Dataverse, where an update changes a field mapping and grounded answers become stale.

B:

This design would make sense if a blocked invoice triggers a hold on related procurement actions, particularly where a network fault leaves the two apps inconsistent; the requested architecture instead needs the dependency that integration tests protect virtual-entity schema and source semantics because an update changes a field mapping and grounded answers become stale.

C:

A replaced model package introduces unknown behavior; the alternate design can address that situation because artifact provenance and integrity checks guard model supply chain, whereas this question needs a solution to an update changes a field mapping and grounded answers become stale.

Agentic AI Business Solutions Architect (AB-100) Practice Tests