Microsoft AB-100 (Agentic AI Business Solutions Architect) Practice Test 21

Question 200

Risk reviewers need a control, not merely a monitoring dashboard. The use case is: Procurement exceptions require comparing several policies and records. Potential harm: A reasoning agent might continue invoking tools without evidence or budget limits. Which protection fits?

  1. Cap tool depth and require evidence for consequential conclusions
  2. Limit deployment to approved users and surfaces
  3. Store source text and flag low-confidence legal translations
  4. Require explicit consent for consequential create or update operations

Explanation

Correct Answer

A:

When a reasoning agent might continue invoking tools without evidence or budget limits, the use case procurement exceptions require comparing several policies and records requires a guardrail aligned with the behavior that reasoning can plan through evidence and tools within enforced boundaries.

Incorrect Answers

B:

Where agent experience must appear in selected Microsoft 365 contexts, the guard can address a Teams-targeted agent package might be made available on unapproved hosts; the exposure that a reasoning agent might continue invoking tools without evidence or budget limits remains relevant to the present assessment.

C:

Another process could need this guardrail because a translated legal clause might lose meaning without access to the original text; the design being assessed supports procurement exceptions require comparing several policies and records and has the explicit exposure that a reasoning agent might continue invoking tools without evidence or budget limits.

D:

The alternate boundary mitigates a plugin might create business records without user consent or approval, which is relevant to a Teams user should query and update a supported external service; it leaves the current workflow vulnerable to a reasoning agent might continue invoking tools without evidence or budget limits.

Question 201

Security and business owners want to keep this capability: A third-party AI tool changes where submitted documents are processed. They must also prevent the following: A third-party processor might move confidential documents without contractual approval. Which guardrail should the design incorporate?

  1. Keep enough prior approved artifacts for recovery
  2. Block deployment pending approved handling and residency evidence
  3. Keep append-only records and redact unnecessary personal details
  4. Require a tested human escalation path for unmatched cases

Explanation

Correct Answer

B:

Because a third-party processor might move confidential documents without contractual approval, a guardrail must constrain the actual system behavior: contract and privacy assessment cover service location and subprocessor use; protecting a response alone cannot enforce that boundary.

Incorrect Answers

A:

This restriction is associated with a preprocessing release changes category labels incorrectly and addresses the exposure that a rollback might restore a model but not its required preprocessing artifacts; it does not specifically protect the target workflow against a third-party processor might move confidential documents without contractual approval.

C:

Where a tool deletes a record following an ambiguous conversation, the guard can address a privileged agent mutation might lack the initiating user and approval trail; the exposure that a third-party processor might move confidential documents without contractual approval remains relevant to the present assessment.

D:

The alternate guard addresses a new support agent might route urgent cases to a general queue for a support agent changes how it classifies priority intents; the present threat, a third-party processor might move confidential documents without contractual approval, needs a different enforcement boundary.

Question 202

One enterprise rollout has 3 separately testable requirements. (1) Business use: A classifier underperforms on one language group. Exposure: An unfair subgroup outcome might remain unresolved at promotion. (2) Business use: Teams promote a custom model because it feels better. Exposure: A candidate might pass overall accuracy while failing a mandatory critical metric. (3) Business use: A document model performs well overall but fails on tax identifiers. Exposure: An uncertain extracted tax ID might flow directly into a posted finance record. Which 3 safeguards directly satisfy these stated cases? Choose THREE.

  1. Block promotion when any mandatory critical threshold fails
  2. Require mitigation before deployment when disparity is material
  3. Route low-confidence tax values for human validation
  4. Do not allow model confidence prose to override missing evidence

Explanation

Correct Answers

A:

For the exposure that a candidate might pass overall accuracy while failing a mandatory critical metric, predefined gates weigh accuracy, latency and business risk explains the protection required when teams promote a custom model because it feels better; this does not replace control of an unfair subgroup outcome might remain unresolved at promotion; an uncertain extracted tax ID might flow directly into a posted finance record.

B:

For the exposure that an unfair subgroup outcome might remain unresolved at promotion, segmented evaluations reveal uneven model behavior explains the protection required when a classifier underperforms on one language group; this does not replace control of a candidate might pass overall accuracy while failing a mandatory critical metric; an uncertain extracted tax ID might flow directly into a posted finance record.

C:

For the exposure that an uncertain extracted tax ID might flow directly into a posted finance record, stratified field metrics protect high-impact values explains the protection required when a document model performs well overall but fails on tax identifiers; this does not replace control of an unfair subgroup outcome might remain unresolved at promotion; a candidate might pass overall accuracy while failing a mandatory critical metric.

Incorrect Answers

D:

This security safeguard addresses the unrelated exposure that confident model language might override an explicit evidence requirement in a workflow where policy answers must be tied to a current source; the question instead names the specific risks an unfair subgroup outcome might remain unresolved at promotion; a candidate might pass overall accuracy while failing a mandatory critical metric; an uncertain extracted tax ID might flow directly into a posted finance record; this choice does not directly mitigate those specific exposures.

Question 203

One enterprise rollout has 2 separately testable requirements. (1) Business use: Users must remain in a proprietary work-order application. Known defect: Requiring a separate chat portal increases context switching. (2) Business use: Staff need a role-specific conversation within Copilot and Teams. Known defect: A separate application breaks established productivity workflows. Which 2 regression tests directly satisfy these stated cases? Choose TWO.

  1. Delay a specialist and confirm timeout and retry behavior
  2. Validate critical user journeys in a supported product trial
  3. Install the agent for a pilot audience and test supported experiences
  4. Execute the target action without leaving the work-order screen

Explanation

Correct Answers

C:

For the business workflow where staff need a role-specific conversation within Copilot and Teams, installing the agent for a pilot audience and testing supported experiences challenges the failure that a separate application breaks established productivity workflows; the release also needs a separate check for users must remain in a proprietary work-order application.

D:

For the business workflow where users must remain in a proprietary work-order application, executing the target action without leaving the work-order screen challenges the failure that requiring a separate chat portal increases context switching; the release also needs a separate check for staff need a role-specific conversation within Copilot and Teams.

Incorrect Answers

A:

This regression test is associated with the different failure where orchestrator returns a partial result as though the workflow finished in a workflow where downstream research agents run longer than the interactive request; it does not directly solve the listed conditions: requiring a separate chat portal increases context switching; a separate application breaks established productivity workflows.

B:

This regression test is associated with the different failure where a bespoke solution creates needless infrastructure overhead in a workflow where the requirement is common and product coverage is adequate; it does not directly solve the listed conditions: requiring a separate chat portal increases context switching; a separate application breaks established productivity workflows.

Question 204

A risk analysis highlights a threat in this workflow: Production components should not be edited casually. Exposure: Direct production customization might diverge from the approved solution. Which enforceable safeguard addresses it at the relevant boundary?

  1. Require finance process-owner approval for production changes
  2. Require a tested human escalation path for unmatched cases
  3. Revoke shared credentials and forbid cross-stage access
  4. Limit production edit roles and monitor drift

Explanation

Correct Answer

D:

Avoiding the consequence where direct production customization might diverge from the approved solution requires a control for production components should not be edited casually; in this architecture, managed solution deployment supports controlled downstream customization boundaries, which is the point at which the boundary can be applied.

Incorrect Answers

A:

For the use case where invoice posting agents affect critical close processes, the control addresses the threat that a finance Copilot prompt might change during a protected close window; the specified agent must be protected against direct production customization might diverge from the approved solution.

B:

The alternate boundary mitigates a new support agent might route urgent cases to a general queue, which is relevant to a support agent changes how it classifies priority intents; it leaves the current workflow vulnerable to direct production customization might diverge from the approved solution.

C:

Test automation must not reach live customer records; its guardrail addresses the possibility that staging credentials might allow writes to production business systems; the present architecture must constrain the different threat that direct production customization might diverge from the approved solution.

Question 205

A solution architecture is being reviewed for this business situation: A preprocessing release changes category labels incorrectly. The observed defect is that restoring code alone does not restore previous predictions. Which architectural decision addresses the underlying requirement?

  1. Register each trained model with dataset and evaluation provenance
  2. Retain rollback snapshots for models and data transformations
  3. Version agent instructions, tool definitions and model references together
  4. Treat index schema and embeddings as versioned artifacts

Explanation

Correct Answer

B:

Without the intended behavior, restoring code alone does not restore previous predictions; the proposed architecture establishes the specific design principle that versioned data transformations and model artifacts enable consistent rollback for the process where a preprocessing release changes category labels incorrectly.

Incorrect Answers

A:

When data scientists cannot explain why a newer classifier behaves differently, a model registry ties versions to lineage, evaluations and approval; in the stated workflow, however, restoring code alone does not restore previous predictions; the relevant architectural behavior is that versioned data transformations and model artifacts enable consistent rollback.

C:

The technical behavior of this alternative is that a complete agent release manifest pins related runtime artifacts; the user case being assessed is a preprocessing release changes category labels incorrectly with a different failure: restoring code alone does not restore previous predictions.

D:

For a search index changes embedding model and chunking approach, this other approach relies on the fact that explicit index versions support compatible retrieval and rollback; the issue here is restoring code alone does not restore previous predictions; that mechanism does not establish the required versioned data transformations and model artifacts enable consistent rollback.

Question 206

A pilot scorecard needs evidence of business improvement rather than only invocation counts. The use case is: A hosted agent gains an additional research tool. Known defect: New planning behavior occasionally leaks internal notes. Which KPI should be selected?

  1. Foundry agent releases with reproducible manifests
  2. Outbound regulated messages with required approvals
  3. Unauthorized finance postings in release tests
  4. Release candidates satisfying critical evaluation thresholds

Explanation

Correct Answer

D:

Because evaluation suites validate quality and safety before publication, the outcome worth following is release candidates satisfying critical evaluation thresholds; the metric is tied to the reported issue—new planning behavior occasionally leaks internal notes—for the scenario where a hosted agent gains an additional research tool.

Incorrect Answers

A:

The premise of this indicator is a complete agent release manifest pins related runtime artifacts; the architecture being evaluated instead serves a hosted agent gains an additional research tool, and its success is measured by release candidates satisfying critical evaluation thresholds.

B:

Outbound regulated messages with required approvals; that indicator describes a distinct workflow where service agents draft customer messages but cannot send without review; it does not quantify release candidates satisfying critical evaluation thresholds, which is needed for the stated case.

C:

The alternative tracks unauthorized finance postings in release tests, arising from the mechanism where authorization regression gates verify posting authority remains constrained; its movement alone says little about the failure that new planning behavior occasionally leaks internal notes.

Question 207

Business leaders reject an activity-only metric for this process: Model and tool changes are promoted independently. They need to detect the issue where a release uses stale prompts against a new tool signature. Which performance indicator fits?

  1. Successful indirect prompt-injection attacks
  2. Foundry agent releases with reproducible manifests
  3. Critical releases with reproducible evaluation and sign-off records
  4. Release candidates satisfying critical evaluation thresholds
  5. Customer handoffs free from integration mismatches

Explanation

Correct Answer

B:

The use case involves model and tool changes are promoted independently and the baseline failure that a release uses stale prompts against a new tool signature; the KPI tracks Foundry agent releases with reproducible manifests, so improvement has an interpretable operational meaning.

Incorrect Answers

A:

External knowledge contains embedded role-change commands; its intended measurement is successful indirect prompt-injection attacks; substituting that result here would not capture whether a release uses stale prompts against a new tool signature has been corrected.

C:

The premise of this indicator is versioned evaluation sets, model IDs and thresholds document release claims; the architecture being evaluated instead serves model and tool changes are promoted independently, and its success is measured by Foundry agent releases with reproducible manifests.

D:

For the workflow where a hosted agent gains an additional research tool, release candidates satisfying critical evaluation thresholds is an actionable measure; it does not establish whether a release uses stale prompts against a new tool signature in the situation under review.

E:

Customer handoffs free from integration mismatches; that indicator describes a distinct workflow where a customer handoff depends on shared record and term mappings; it does not quantify Foundry agent releases with reproducible manifests, which is needed for the stated case.

Question 208

Governance approves the business purpose but requires control of this exposure: Staging credentials might allow writes to production business systems. The agent will be used when test automation must not reach live customer records. Which boundary is most defensible?

  1. Limit production edit roles and monitor drift
  2. Revoke shared credentials and forbid cross-stage access
  3. Roll back or withhold publication if acceptance checks fail
  4. Scope the identity to only required resource permissions

Explanation

Correct Answer

B:

When staging credentials might allow writes to production business systems, the use case test automation must not reach live customer records requires a guardrail aligned with the behavior that environment-scoped identities isolate access and simplify incident containment.

Incorrect Answers

A:

Another process could need this guardrail because direct production customization might diverge from the approved solution; the design being assessed supports test automation must not reach live customer records and has the explicit exposure that staging credentials might allow writes to production business systems.

C:

The safeguard for solution import succeeds but users still see the previous behavior deals with a different threat: an imported but unpublished agent might cause users to see stale behavior; even after it is enforced, the exposure that staging credentials might allow writes to production business systems remains possible in the target workflow where environment-scoped identities isolate access and simplify incident containment.

D:

The alternate guard addresses long-lived credentials might grant an agent unnecessary Azure resource privileges for a backend agent must call an approved Azure service without embedded secrets; the present threat, staging credentials might allow writes to production business systems, needs a different enforcement boundary.

Question 209

During integration testing, a model upgrade breaks downstream response formatting. The associated business context is: Business applications depend on the agent endpoint continuously. Which architectural remedy is the strongest match?

  1. Use progressive deployment and rollback for a stable agent endpoint
  2. Maintain an enterprise inventory of production agents and owners
  3. Integrate runtime tracing and alerting into release readiness
  4. Promote grounding datasets through quality gates

Explanation

Correct Answer

A:

Because canary traffic and versioned rollback reduce release risk, this approach corrects the reported breakdown where a model upgrade breaks downstream response formatting; the affected situation is that business applications depend on the agent endpoint continuously.

Incorrect Answers

B:

This design would make sense if business units create agents with no central visibility, particularly where security cannot identify which agents use protected connectors; the requested architecture instead needs the dependency that canary traffic and versioned rollback reduce release risk because a model upgrade breaks downstream response formatting.

C:

The alternative addresses the on-call team cannot explain failed tool calls by a design in which connected observability supplies traces and operational metrics; it does not supply the missing relationship for business applications depend on the agent endpoint continuously, where a model upgrade breaks downstream response formatting.

D:

For new documents enter a production source automatically, this other approach relies on the fact that ingestion gates check classification, freshness and ownership; the issue here is a model upgrade breaks downstream response formatting; that mechanism does not establish the required canary traffic and versioned rollback reduce release risk.

Question 210

Implementation alternatives must be compared for a workflow where the model submits malformed identifiers to an operational API. The pilot shows this gap: Invalid record IDs produce updates to the wrong business object. Which alternative should lead the solution?

  1. Build a bespoke agent when differentiated orchestration is essential
  2. Use staged rollout with feature flags and pilot cohorts
  3. Validate generated parameters against typed business schemas
  4. Require approval before irreversible agent actions

Explanation

Correct Answer

C:

Because typed inputs and server validation reject malformed or out-of-scope calls, this approach corrects the reported breakdown where invalid record IDs produce updates to the wrong business object; the affected situation is that the model submits malformed identifiers to an operational API.

Incorrect Answers

A:

There is a valid place for the mechanism that custom engineering trades more control for more operating responsibility, namely competitive requirements depend on unique multi-model control; it does not remove the specific defect that invalid record IDs produce updates to the wrong business object from this case.

B:

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 invalid record IDs produce updates to the wrong business object; that mechanism does not establish the required typed inputs and server validation reject malformed or out-of-scope calls.

D:

For another type of architecture, confirmation and human approval separate advice from execution authority is useful; this design must instead support the model submits malformed identifiers to an operational API despite the problem where invalid record IDs produce updates to the wrong business object.

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