Microsoft AB-100 (Agentic AI Business Solutions Architect) Practice Test 108

Question 1070

Release reviewers will not accept a generic success message. The business scenario is that a research agent repeatedly queries sources on difficult cases, and the historical failure is: A single case exhausts model and tool quotas. Which test is appropriately targeted?

  1. Test borderline utterances against both descriptions
  2. Ask unrelated questions and verify no supplier action is triggered
  3. Inject an unresolvable task and check budget termination
  4. Evaluate held-out rare labels and previously unseen layouts

Explanation

Correct Answer

C:

The scenario is that a research agent repeatedly queries sources on difficult cases; the defect to rule out is that a single case exhausts model and tool quotas; testing via injecting an unresolvable task and checking budget termination challenges the design dependency where time, step and spend budgets terminate unproductive loops.

Incorrect Answers

A:

The suggested test serves two topics claim nearly identical scheduling intents, in which agent invokes the wrong booking action for a reschedule; the present scenario instead concerns the reported problem where a single case exhausts model and tool quotas.

B:

For a bot only needs to check warranty and prepare a draft, the alternate test is about whether limited intent and tool scope prevent unnecessary actions; it cannot prove the outcome needed after the reported regression where a single case exhausts model and tool quotas.

D:

The verification concerning general models repeatedly misread rare engineering labels checks domain-trained models can be evaluated when generic baselines fall short against prompt changes fail to resolve a measured specialist error; its result says little about the observed issue where a single case exhausts model and tool quotas.

Question 1071

A quality engineer is writing a regression case after the incident where agent-generated pages render an unsafe instruction or link. The underlying business scenario is: A tool response includes HTML or links supplied by third parties. What should the test exercise?

  1. Reproduce a reported answer from recorded non-sensitive context
  2. Rebuild a staged agent from recorded versions and run smoke tests
  3. Feed adversarial markup in a synthetic tool response
  4. Run the same corpus in a CI release candidate
  5. Compare controlled tests before and after tool curation

Explanation

Correct Answer

C:

Agent-generated pages render an unsafe instruction or link; a release test must establish that this no longer occurs; feeding adversarial markup in a synthetic tool response measures the specific behavior for which trusted rendering and output validation constrain returned content.

Incorrect Answers

A:

Reproducing a reported answer from recorded non-sensitive context verifies an adjacent objective where users report stale answers after a policy refresh; it leaves untested the causal path in which trusted rendering and output validation constrain returned content for this case.

B:

A pass on that verification would speak to model and tool changes are promoted independently; it would not show that trusted rendering and output validation constrain returned content, which is needed to avoid the observed defect agent-generated pages render an unsafe instruction or link.

D:

A pass on that verification would speak to prompt revisions cause intermittent regressions; it would not show that trusted rendering and output validation constrain returned content, which is needed to avoid the observed defect agent-generated pages render an unsafe instruction or link.

E:

This check exercises a different system behavior: clear capabilities and restricted sets improve selection reliability; a pass there cannot establish that agent-generated pages render an unsafe instruction or link no longer occurs in the required workflow.

Question 1072

The operating team must monitor whether an architectural correction works. Business context: Deployment automation pulls model artifacts from unverified origins. Historical symptom: A replaced model package introduces unknown behavior. Which outcome measure is the best match?

  1. Model deployments passing artifact-integrity controls
  2. Deployed models with complete registry evidence
  3. Unauthorized accesses to model-training datasets
  4. Model changes with complete actor and approval records

Explanation

Correct Answer

A:

For the business use where deployment automation pulls model artifacts from unverified origins, the result to quantify is model deployments passing artifact-integrity controls; it connects measurement to the mechanism where artifact provenance and integrity checks guard model supply chain, not merely to agent traffic.

Incorrect Answers

B:

A change in deployed models with complete registry evidence is informative for data scientists cannot explain why a newer classifier behaves differently; it does not establish whether the current process achieves model deployments passing artifact-integrity controls.

C:

The premise of this indicator is segregated datasets and access controls protect sensitive data; the architecture being evaluated instead serves deployment automation pulls model artifacts from unverified origins, and its success is measured by model deployments passing artifact-integrity controls.

D:

The alternative tracks model changes with complete actor and approval records, arising from the mechanism where deployment audit links actor, approved artifact and release time; its movement alone says little about the failure that a replaced model package introduces unknown behavior.

Question 1073

One enterprise rollout has 2 separately testable requirements. (1) Business use: A bot only needs to check warranty and prepare a draft. Known defect: It autonomously contacts a supplier without approval. (2) Business use: Network retries can repeat a single purchase request. Known defect: Two purchase orders appear from one user intent. Which 2 diagnostic procedures directly satisfy these stated cases? Choose TWO.

  1. Compare advertised tools with backend permissions
  2. Compare tool inventory with legitimate workflow steps
  3. Inspect duplicate request IDs across retries
  4. Analyze containment failures and repeated intent fallback patterns

Explanation

Correct Answers

B:

Comparing tool inventory with legitimate workflow steps can reveal why it autonomously contacts a supplier without approval when a bot only needs to check warranty and prepare a draft; that finding is needed alongside the independent diagnosis for network retries can repeat a single purchase request.

C:

Inspecting duplicate request IDs across retries can reveal why two purchase orders appear from one user intent when network retries can repeat a single purchase request; that finding is needed alongside the independent diagnosis for a bot only needs to check warranty and prepare a draft.

Incorrect Answers

A:

This diagnosis is associated with the different failure where each agent builds a different unaudited HTTP wrapper in a workflow where several agents need the same inventory query operation; it does not directly solve the listed conditions: it autonomously contacts a supplier without approval; two purchase orders appear from one user intent.

D:

This diagnosis is associated with the different failure where repeated fallback messages trap callers in a loop in a workflow where contact Center agents sometimes cannot resolve complex billing intent; it does not directly solve the listed conditions: it autonomously contacts a supplier without approval; two purchase orders appear from one user intent.

Question 1074

After deployment, stakeholders need an interpretable measure for this use case: Retrieved policy fragments lose context when documents are split. The pilot showed that answers cite isolated sentences without section qualifiers. Which metric captures the needed result?

  1. Holdout error rate by business-critical segment
  2. Grounding recall for critical enterprise questions
  3. Answer correctness on cross-section retrieval cases
  4. Unauthorized data fields sent to model actions

Explanation

Correct Answer

C:

The observed shortcoming is that answers cite isolated sentences without section qualifiers; measuring answer correctness on cross-section retrieval cases gives business owners a directly relevant indicator for the process where retrieved policy fragments lose context when documents are split.

Incorrect Answers

A:

The alternative tracks holdout error rate by business-critical segment, arising from the mechanism where separation of training and evaluation yields realistic generalization checks; its movement alone says little about the failure that answers cite isolated sentences without section qualifiers.

B:

For the workflow where stakeholders report hallucinations but cannot isolate search errors, grounding recall for critical enterprise questions is an actionable measure; it does not establish whether answers cite isolated sentences without section qualifiers in the situation under review.

D:

The premise of this indicator is server-side authorization and minimized inputs protect data flow; the architecture being evaluated instead serves retrieved policy fragments lose context when documents are split, and its success is measured by answer correctness on cross-section retrieval cases.

Question 1075

A risk analysis highlights a threat in this workflow: A tool response includes HTML or links supplied by third parties. Exposure: Untrusted HTML in a tool result might be rendered as executable content. Which enforceable safeguard addresses it at the relevant boundary?

  1. Reject missing, invalid or out-of-scope authorization tokens
  2. Remove unused privileged actions from the agent
  3. Encode untrusted content and separate data from instructions
  4. Fail closed on authorization and schema-validation failures

Explanation

Correct Answer

C:

In the context where a tool response includes HTML or links supplied by third parties, the exposure is that untrusted HTML in a tool result might be rendered as executable content; containing it requires a guardrail in the operation where trusted rendering and output validation constrain returned content.

Incorrect Answers

A:

That protection is tied to the mechanism that server-side access checks enforce real actor entitlements; it does not prevent the distinct risk where untrusted HTML in a tool result might be rendered as executable content in the system under review.

B:

This restriction is associated with planner calls the wrong tool among overlapping capabilities and addresses the exposure that a planner might choose an unused privileged tool that remains exposed; it does not specifically protect the target workflow against untrusted HTML in a tool result might be rendered as executable content.

D:

The proposed control targets malformed calls might pass a happy-path suite yet mutate records, while the given risk scenario is untrusted HTML in a tool result might be rendered as executable content; these exposures arise in different operational contexts.

Question 1076

A risk analysis highlights a threat in this workflow: Users must remain in a proprietary work-order application. Exposure: A hidden app field could still be retrieved through an over-privileged backend call. Which enforceable safeguard addresses it at the relevant boundary?

  1. Enforce backend permissions separately from frontend display
  2. Prevent duplicate side effects during task retries
  3. Narrow allowed tool capabilities and verify server-side authorization
  4. Review tenancy, data terms, exit and support commitments

Explanation

Correct Answer

A:

Because a hidden app field could still be retrieved through an over-privileged backend call, a guardrail must constrain the actual system behavior: app-hosted integration brings relevant AI actions into an owned user interface; protecting a response alone cannot enforce that boundary.

Incorrect Answers

B:

For the use case where downstream research agents run longer than the interactive request, the control addresses the threat that retried delegated work could charge a customer or create a record more than once; the specified agent must be protected against a hidden app field could still be retrieved through an over-privileged backend call.

C:

This restriction is associated with a Copilot Studio agent needs a reusable interface to governed operations and addresses the exposure that a shared MCP server might honor an unauthorized tool invocation; it does not specifically protect the target workflow against a hidden app field could still be retrieved through an over-privileged backend call.

D:

The alternate guard addresses a managed service might fail a required data-processing or exit condition for the requirement is common and product coverage is adequate; the present threat, a hidden app field could still be retrieved through an over-privileged backend call, needs a different enforcement boundary.

Question 1077

Implementation alternatives must be compared for a workflow where a request spans procurement, finance and policy expertise. The pilot shows this gap: One general agent conflates expertise and may call the wrong tools. Which alternative should lead the solution?

  1. Embed a Copilot-powered interaction inside an existing line-of-business app
  2. Use a supervisor agent that routes to specialized child agents
  3. Use A2A for agent-to-agent collaboration across runtimes
  4. Decompose end-to-end workflows into agent-appropriate tasks

Explanation

Correct Answer

B:

The architecture must reconcile two facts: a request spans procurement, finance and policy expertise, yet one general agent conflates expertise and may call the wrong tools; the selected pattern does so through the mechanism that a coordinating agent selects bounded specialists and merges their outputs.

Incorrect Answers

A:

For users must remain in a proprietary work-order application, this other approach relies on the fact that app-hosted integration brings relevant AI actions into an owned user interface; the issue here is one general agent conflates expertise and may call the wrong tools; that mechanism does not establish the required a coordinating agent selects bounded specialists and merges their outputs.

C:

The technical behavior of this alternative is that an agent interoperability protocol supports explicit task delegation between agents; the user case being assessed is a request spans procurement, finance and policy expertise with a different failure: one general agent conflates expertise and may call the wrong tools.

D:

This design would make sense if a business process mixes deterministic calculations and open-ended judgments, particularly where a single agent unpredictably handles policy rules and narrative summarization; the requested architecture instead needs the dependency that a coordinating agent selects bounded specialists and merges their outputs because one general agent conflates expertise and may call the wrong tools.

Question 1078

A product owner must measure whether an AI-enabled change actually improves this process: Work can proceed from events without human initiation but needs bounded autonomy. The earlier problem was: A background agent has triggered irreversible duplicate actions. Which outcome indicator is most relevant?

  1. Autonomous tasks completed without unauthorized changes
  2. Percentage of delegated tasks completed within agreed deadlines
  3. Percentage of material analytical claims with checked citations
  4. Time from source exception to downstream acknowledgement

Explanation

Correct Answer

A:

The observed shortcoming is that a background agent has triggered irreversible duplicate actions; measuring autonomous tasks completed without unauthorized changes gives business owners a directly relevant indicator for the process where work can proceed from events without human initiation but needs bounded autonomy.

Incorrect Answers

B:

The alternative tracks percentage of delegated tasks completed within agreed deadlines, arising from the mechanism where task IDs, status transitions, deadlines and durable handoff state support coordination; its movement alone says little about the failure that a background agent has triggered irreversible duplicate actions.

C:

For the workflow where analysts compare unstructured reports but must retain human interpretation, percentage of material analytical claims with checked citations is an actionable measure; it does not establish whether a background agent has triggered irreversible duplicate actions in the situation under review.

D:

That scorecard concerns the activity in which invoice holds must trigger exception handling without polling and its result of time from source exception to downstream acknowledgement; the reported shortcoming here is a background agent has triggered irreversible duplicate actions.

Question 1079

A risk analysis highlights a threat in this workflow: Model bills increase although request volume is stable. Exposure: A prompt might accumulate costly context with no consumption guardrail. Which enforceable safeguard addresses it at the relevant boundary?

  1. Roll back if privileged actions appear in unexpected paths
  2. Reject mismatched schema/model combinations
  3. Set cost budgets and investigate anomalous context expansion
  4. Roll back a routing change when a critical segment regresses

Explanation

Correct Answer

C:

In the scenario where model bills increase although request volume is stable, the explicit risk is that a prompt might accumulate costly context with no consumption guardrail; enforcement must follow the mechanism by which token telemetry isolates prompt, retrieval and generation cost changes.

Incorrect Answers

A:

The safeguard for a new topic description changes which agent is invoked deals with a different threat: a new orchestration release might suddenly include unexpected privileged tool paths; even after it is enforced, the exposure that a prompt might accumulate costly context with no consumption guardrail remains possible in the target workflow where token telemetry isolates prompt, retrieval and generation cost changes.

B:

That protection is tied to the mechanism that bundled preprocessing, model and label maps maintain contract consistency; it does not prevent the distinct risk where a prompt might accumulate costly context with no consumption guardrail in the system under review.

D:

This restriction is associated with average accuracy hides failures on complex finance questions and addresses the exposure that a model router change might silently degrade the highest-risk task segment; it does not specifically protect the target workflow against a prompt might accumulate costly context with no consumption guardrail.

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?

  1. Require heightened review for autonomous privileged agents
  2. Retain audit records of source replacements and withdrawals
  3. Reject cached results that omit original security context
  4. Do not allow model confidence prose to override missing evidence
  5. 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.

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