Microsoft AB-100 (Agentic AI Business Solutions Architect) Practice Test 4

Question 30

Governance approves the business purpose but requires control of this exposure: An unapproved change to the approval matrix could silently alter authorization decisions. The agent will be used when the task follows fixed approval branches with no interpretation needed. Which boundary is most defensible?

  1. Deny unapproved tools regardless of model-generated arguments
  2. Reject malformed output before writing business records
  3. Validate sending-agent identity and delegated authorization
  4. Require policy-owner approval for any rule change

Explanation

Correct Answer

D:

The task follows fixed approval branches with no interpretation needed; its risk scenario is that an unapproved change to the approval matrix could silently alter authorization decisions; the safeguard fits because explicit conditions execute known business rules predictably and leave an auditable branch history and the restriction addresses that specific exposure.

Incorrect Answers

A:

The alternate boundary mitigates planner-generated arguments might invoke tools that the user was never permitted to call, which is relevant to agent instructions mention many APIs but only two are permitted; it leaves the current workflow vulnerable to an unapproved change to the approval matrix could silently alter authorization decisions.

B:

The alternate boundary mitigates invalid generated JSON might be written directly into a business record, which is relevant to agents return inconsistent fields for downstream integrations; it leaves the current workflow vulnerable to an unapproved change to the approval matrix could silently alter authorization decisions.

C:

That protection is tied to the mechanism that an agent interoperability protocol supports explicit task delegation between agents; it does not prevent the distinct risk where an unapproved change to the approval matrix could silently alter authorization decisions in the system under review.

Question 31

An enterprise team is making a platform design decision. Business context: Agent instructions mention many APIs but only two are permitted. Present failure: The agent can invoke an unintended high-privilege operation. Which approach resolves the mismatch?

  1. Give prompts explicit role, task, constraints and response schema
  2. Validate generated parameters against typed business schemas
  3. Use staged rollout with feature flags and pilot cohorts
  4. Constrain agent tools by explicit allowlists and permissions

Explanation

Correct Answer

D:

Agent instructions mention many APIs but only two are permitted; its failing behavior is that the agent can invoke an unintended high-privilege operation; applying the underlying design principle—allowlisting and server-enforced identity restrict actual executable capability—connects the requirement to an implementable control.

Incorrect Answers

A:

Narrative answers cannot be parsed by consuming workflows; the alternate design can address that situation because clear instructions and structured outputs reduce avoidable ambiguity, whereas this question needs a solution to the agent can invoke an unintended high-privilege operation.

B:

The alternative addresses invalid record IDs produce updates to the wrong business object by a design in which typed inputs and server validation reject malformed or out-of-scope calls; it does not supply the missing relationship for agent instructions mention many APIs but only two are permitted, where the agent can invoke an unintended high-privilege operation.

C:

For new behavior must be evaluated without exposing the full workforce, this other approach relies on the fact that controlled release gates restrict blast radius and support rollback; the issue here is the agent can invoke an unintended high-privilege operation; that mechanism does not establish the required allowlisting and server-enforced identity restrict actual executable capability.

Question 32

A policy review covers an agent for which product catalog queries mix synonyms and exact serial codes. Its most relevant exposure is that a reranker might return a relevant passage that the user cannot access. What safeguard should be required before release?

  1. Protect recordings and redact sensitive transcript elements
  2. Review low-confidence financial fields before posting
  3. Validate source access after ranking and before response
  4. Prevent deployment of unapproved model/data combinations

Explanation

Correct Answer

C:

In the context where product catalog queries mix synonyms and exact serial codes, the exposure is that a reranker might return a relevant passage that the user cannot access; containing it requires a guardrail in the operation where hybrid ranking combines lexical precision and semantic relevance.

Incorrect Answers

A:

For the use case where quality teams need searchable insights from recorded calls, the control addresses the threat that recorded customer speech or transcript fields might be exposed to unintended viewers; the specified agent must be protected against a reranker might return a relevant passage that the user cannot access.

B:

The safeguard for vendors send variable-layout invoices and shipping documents deals with a different threat: uncertain tax identifiers extracted from invoices might be posted without review; even after it is enforced, the exposure that a reranker might return a relevant passage that the user cannot access remains possible in the target workflow where hybrid ranking combines lexical precision and semantic relevance.

D:

Where retraining uses revised labels and a different source snapshot, the guard can address a model release might combine training data with an unapproved candidate artifact; the exposure that a reranker might return a relevant passage that the user cannot access remains relevant to the present assessment.

Question 33

An investment review asks what measurable result would justify continuing an AI solution where shipment exceptions should initiate investigation without a chat message. The original issue is: Teams learn about problems only when someone asks the agent. Which indicator should be tracked?

  1. Agent runs terminating within predefined budgets
  2. Unauthorized ancillary actions per task
  3. Eligible exception events creating one investigation
  4. Analyst-accepted explanations with verified references

Explanation

Correct Answer

C:

For the use case where shipment exceptions should initiate investigation without a chat message, the indicator eligible exception events creating one investigation is worth baselining against the earlier issue where teams learn about problems only when someone asks the agent.

Incorrect Answers

A:

For the workflow where a research agent repeatedly queries sources on difficult cases, agent runs terminating within predefined budgets is an actionable measure; it does not establish whether teams learn about problems only when someone asks the agent in the situation under review.

B:

A bot only needs to check warranty and prepare a draft; its intended measurement is unauthorized ancillary actions per task; substituting that result here would not capture whether teams learn about problems only when someone asks the agent has been corrected.

D:

For the workflow where finance asks why a variance was flagged, analyst-accepted explanations with verified references is an actionable measure; it does not establish whether teams learn about problems only when someone asks the agent in the situation under review.

Question 34

A policy review covers an agent for which an agent scales from a trial to enterprise usage. Its most relevant exposure is that increased model traffic might run beyond the project budget without alerts. What safeguard should be required before release?

  1. Keep a known-good model and agent configuration for rollback
  2. Set usage budgets, cost alerts and routing limits
  3. Keep critical numeric calculations in tested services
  4. Add human approval for consequential recommendations

Explanation

Correct Answer

B:

In the context where an agent scales from a trial to enterprise usage, the exposure is that increased model traffic might run beyond the project budget without alerts; containing it requires a guardrail in the operation where workload forecasts include model usage, connectors, storage and observability.

Incorrect Answers

A:

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 increased model traffic might run beyond the project budget without alerts remains possible in the target workflow where workload forecasts include model usage, connectors, storage and observability.

C:

The alternate boundary mitigates probabilistic generation might replace authoritative arithmetic used in financial controls, which is relevant to a business process mixes deterministic calculations and open-ended judgments; it leaves the current workflow vulnerable to increased model traffic might run beyond the project budget without alerts.

D:

The proposed control targets an automated recommendation might be executed without human accountability, while the given risk scenario is increased model traffic might run beyond the project budget without alerts; these exposures arise in different operational contexts.

Question 35

Before production promotion, the team needs evidence for this setting: Business, platform, security and legal teams make conflicting AI decisions. A prior regression caused the following: Deployments stall because approvals and accountability are unclear. Which validation is most persuasive?

  1. Run a hypothetical high-risk deployment through the decision-rights matrix
  2. Attempt low-risk and high-risk actions and inspect approval enforcement
  3. Route a pilot cohort to the new version and compare outcomes
  4. Update a case and verify an authorized sales workflow sees it

Explanation

Correct Answer

A:

For the required workflow where business, platform, security and legal teams make conflicting AI decisions, the test needs to reproduce or prevent the failure where deployments stall because approvals and accountability are unclear; running a hypothetical high-risk deployment through the decision-rights matrix exercises the path where a named strategy team assigns decision rights and resolves cross-function tradeoffs.

Incorrect Answers

B:

Although the exercise might catch the same agent can read an invoice and commit a payment unchecked, it would not directly validate the mechanism that a named strategy team assigns decision rights and resolves cross-function tradeoffs in the target implementation.

C:

The verification concerning new behavior must be evaluated without exposing the full workforce checks controlled release gates restrict blast radius and support rollback against a prompt update changes decisions for all users at once; its result says little about the observed issue where deployments stall because approvals and accountability are unclear.

D:

This check exercises a different system behavior: shared authorized records support coordinated business context; a pass there cannot establish that deployments stall because approvals and accountability are unclear no longer occurs in the required workflow.

Question 36

A business stakeholder describes this operating situation: A summarizer sends output into a ticket ingestion service. Testing instead shows that free-form paragraphs break required schema parsing. What design approach is most appropriate?

  1. Configure contextual customer-response drafting with policy sources
  2. Use a structured response contract for downstream consumption
  3. Use an agent flow for repeatable ordered business steps
  4. Build a scoped task agent for repeatable knowledge-to-action work
  5. Control creativity for regulated response templates

Explanation

Correct Answer

B:

A summarizer sends output into a ticket ingestion service; its failing behavior is that free-form paragraphs break required schema parsing; applying the underlying design principle—structured outputs define required fields and allowable values—connects the requirement to an implementable control.

Incorrect Answers

A:

Replies cite generic guidance instead of contract-specific policy; the alternate design can address that situation because contextual grounding provides relevant policy and customer facts, whereas this question needs a solution to free-form paragraphs break required schema parsing.

C:

The alternative addresses an unconstrained planner sends a notification before authorization by a design in which defined flow steps make sequencing and error paths explicit; it does not supply the missing relationship for a summarizer sends output into a ticket ingestion service, where free-form paragraphs break required schema parsing.

D:

For another type of architecture, single-purpose tools and typed parameters bound the task is useful; this design must instead support a summarizer sends output into a ticket ingestion service despite the problem where free-form paragraphs break required schema parsing.

E:

For compliance letters must preserve approved language and disclaimers, this other approach relies on the fact that templating and approved clauses limit generated variation; the issue here is free-form paragraphs break required schema parsing; that mechanism does not establish the required structured outputs define required fields and allowable values.

Question 37

An agent rollout is intended for this situation: An external API changes a required field used by an agent. The known problem is that agent actions fail only after the connector reaches production. Which success measure connects to that specific outcome?

  1. Data-policy violations blocked in integration tests
  2. Retrieval regressions following index promotions
  3. Connector schema regressions caught before production
  4. Unauthorized cross-environment requests blocked

Explanation

Correct Answer

C:

For the use case where an external API changes a required field used by an agent, the indicator connector schema regressions caught before production is worth baselining against the earlier issue where agent actions fail only after the connector reaches production.

Incorrect Answers

A:

For the workflow where a Copilot Studio agent can read corporate data and use public messaging, data-policy violations blocked in integration tests is an actionable measure; it does not establish whether agent actions fail only after the connector reaches production in the situation under review.

B:

A change in retrieval regressions following index promotions is informative for a search index changes embedding model and chunking approach; it does not establish whether the current process achieves connector schema regressions caught before production.

D:

The proposed metric concerns test automation must not reach live customer records, whereas the failure in this case is that agent actions fail only after the connector reaches production; the indicator needed here is connector schema regressions caught before production.

Question 38

Business leaders reject an activity-only metric for this process: Teams propose that an agent perform both low-risk lookups and financial postings. They need to detect the issue where the same agent can read an invoice and commit a payment unchecked. Which performance indicator fits?

  1. Incremental outcome lift relative to control
  2. High-impact actions executed without required approval
  3. Error-rate delta between canary and control cohorts
  4. Unsupported regulated answers reaching users

Explanation

Correct Answer

B:

The use case involves teams propose that an agent perform both low-risk lookups and financial postings and the baseline failure that the same agent can read an invoice and commit a payment unchecked; the KPI tracks high-impact actions executed without required approval, so improvement has an interpretable operational meaning.

Incorrect Answers

A:

Management attributes a sales lift to Copilot without a comparison; its intended measurement is incremental outcome lift relative to control; substituting that result here would not capture whether the same agent can read an invoice and commit a payment unchecked has been corrected.

C:

Although error-rate delta between canary and control cohorts is observable, the affected process involves teams propose that an agent perform both low-risk lookups and financial postings; its acceptance outcome is high-impact actions executed without required approval, not the alternative measurement.

D:

For the workflow where the system cannot always find evidence for regulated guidance, unsupported regulated answers reaching users is an actionable measure; it does not establish whether the same agent can read an invoice and commit a payment unchecked in the situation under review.

Question 39

Before production promotion, the team needs evidence for this setting: The same agent must move from development to test and production. A prior regression caused the following: Manual recreation causes mismatched topics and actions. Which validation is most persuasive?

  1. Force a downstream error and verify cross-system traceability
  2. Exercise both old and new agent builds against staged endpoints
  3. Import a test solution and verify expected agent behavior
  4. Replay representative requests against the new connector version

Explanation

Correct Answer

C:

Importing a test solution and verifying expected agent behavior challenges the path where solution-based ALM transports supported components under controlled releases for the business setting the same agent must move from development to test and production; the specific defect to exclude is that manual recreation causes mismatched topics and actions.

Incorrect Answers

A:

The verification concerning a request crosses Copilot Studio, a flow and the CRM backend checks correlation identifiers link the distributed operational path against logs cannot reconstruct which API mutation followed an answer; its result says little about the observed issue where manual recreation causes mismatched topics and actions.

B:

In the alternate setting where a new action version changes how cases are updated, the test exercises coordinated versioning preserves consumer/provider compatibility; even a pass would not show that manual recreation causes mismatched topics and actions has been prevented.

D:

A pass on that verification would speak to an external API changes a required field used by an agent; it would not show that solution-based ALM transports supported components under controlled releases, which is needed to avoid the observed defect manual recreation causes mismatched topics and actions.

Question 40

A proposed deployment supports this business case: A unique domain vocabulary produces persistent measured errors. Reviewers also warn that customization might promote a model trained on data without a privacy review. Which restriction mitigates that concrete risk?

  1. Keep proprietary calculations and secrets on the service side
  2. Budget for independent hosting, safety testing and monitoring
  3. Avoid publishing benefits without attribution and uncertainty bounds
  4. Apply privacy review and model-version rollback before release

Explanation

Correct Answer

D:

Because customization might promote a model trained on data without a privacy review, a guardrail must constrain the actual system behavior: a measured capability gap, appropriate dataset and MLOps plan justify customization; protecting a response alone cannot enforce that boundary.

Incorrect Answers

A:

The alternate guard addresses a third-party conversational host might receive proprietary scoring logic or secrets for standard conversation needs a proprietary scoring service; the present threat, customization might promote a model trained on data without a privacy review, needs a different enforcement boundary.

B:

The alternate guard addresses a bespoke runtime might be shipped without an accountable security or support team for a workflow needs proprietary scheduling, model choice and runtime control; the present threat, customization might promote a model trained on data without a privacy review, needs a different enforcement boundary.

C:

The alternate guard addresses seasonal demand could be misrepresented as attributable AI benefit for management attributes a sales lift to Copilot without a comparison; the present threat, customization might promote a model trained on data without a privacy review, needs a different enforcement boundary.

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