Microsoft AB-100 (Agentic AI Business Solutions Architect) Practice Test 59

Question 580

One enterprise rollout has 2 separately testable requirements. (1) Business use: Service representatives draft responses from authorized knowledge. Known defect: Replies cite generic guidance instead of contract-specific policy. (2) Business use: A representative must initiate an exception workflow from the case. Known defect: AI suggestions are never converted into traceable escalation records. Which 2 diagnostic procedures directly satisfy these stated cases? Choose TWO.

  1. Compare source timestamps and principal mapping on connector calls
  2. Trace each suggested promise to the customer's entitlement and policy
  3. Verify virtual entity enablement and row-level access
  4. Review action arguments and created escalation ownership

Explanation

Correct Answers

B:

Tracing each suggested promise to the customer's entitlement and policy can reveal why replies cite generic guidance instead of contract-specific policy when service representatives draft responses from authorized knowledge; that finding is needed alongside the independent diagnosis for a representative must initiate an exception workflow from the case.

D:

Reviewing action arguments and created escalation ownership can reveal why AI suggestions are never converted into traceable escalation records when a representative must initiate an exception workflow from the case; that finding is needed alongside the independent diagnosis for service representatives draft responses from authorized knowledge.

Incorrect Answers

A:

This diagnosis is associated with the different failure where Copilot relies on last quarter's manually imported account report in a workflow where sellers need current credit exposure held outside CRM; it does not directly solve the listed conditions: replies cite generic guidance instead of contract-specific policy; AI suggestions are never converted into traceable escalation records.

C:

This diagnosis is associated with the different failure where a static knowledge upload gives stale purchase-order states in a workflow where finance users need current operational data in contextual help; it does not directly solve the listed conditions: replies cite generic guidance instead of contract-specific policy; AI suggestions are never converted into traceable escalation records.

Question 581

A risk analysis highlights a threat in this workflow: Compliance letters must preserve approved language and disclaimers. Exposure: A model might alter mandatory legal wording in customer correspondence. Which enforceable safeguard addresses it at the relevant boundary?

  1. Require human sign-off when regulated language is modified
  2. Provide accessible alternatives and honor channel policy
  3. Reject generated fields outside the approved schema
  4. Validate types and authorization at the called service

Explanation

Correct Answer

A:

Because a model might alter mandatory legal wording in customer correspondence, a guardrail must constrain the actual system behavior: templating and approved clauses limit generated variation; protecting a response alone cannot enforce that boundary.

Incorrect Answers

B:

The alternate boundary mitigates the voice channel might provide an inaccessible or unusable response form, which is relevant to the same agent is offered in voice and asynchronous chat; it leaves the current workflow vulnerable to a model might alter mandatory legal wording in customer correspondence.

C:

The safeguard for an agent must turn service notes into a fixed JSON summary deals with a different threat: generated response keys might be trusted even when they violate the schema; even after it is enforced, the exposure that a model might alter mandatory legal wording in customer correspondence remains possible in the target workflow where templating and approved clauses limit generated variation.

D:

Where the orchestrator chooses between similar invoice and credit tools, the guard can address an API might accept a valid-looking but unauthorized generated identifier; the exposure that a model might alter mandatory legal wording in customer correspondence remains relevant to the present assessment.

Question 582

A solution architecture is being reviewed for this business situation: An agent must turn service notes into a fixed JSON summary. The observed defect is that a general conversational turn produces inconsistent output fields. Which architectural decision addresses the underlying requirement?

  1. Define robust tool descriptions and input schemas for planner selection
  2. Disambiguate overlapping topic descriptions and triggers
  3. Use a declarative agent for role-tailored Microsoft 365 work
  4. Use a prompt action for bounded text transformation

Explanation

Correct Answer

D:

The architecture must reconcile two facts: an agent must turn service notes into a fixed JSON summary, yet a general conversational turn produces inconsistent output fields; the selected pattern does so through the mechanism that prompt actions encapsulate reusable inputs and expected outputs.

Incorrect Answers

A:

The technical behavior of this alternative is that specific tool metadata helps the planner choose and parameterize actions; the user case being assessed is an agent must turn service notes into a fixed JSON summary with a different failure: a general conversational turn produces inconsistent output fields.

B:

The alternative addresses agent invokes the wrong booking action for a reschedule by a design in which distinct descriptions and inputs improve topic selection; it does not supply the missing relationship for an agent must turn service notes into a fixed JSON summary, where a general conversational turn produces inconsistent output fields.

C:

This design would make sense if employees want curated assistance without owning a separate LLM runtime, particularly where a custom service would duplicate built-in host orchestration; the requested architecture instead needs the dependency that prompt actions encapsulate reusable inputs and expected outputs because a general conversational turn produces inconsistent output fields.

Question 583

A release must prove that agent cites a private HR document to unauthorized staff will not recur. The application is used in this situation: Users have different rights to the same knowledge corpus. Which acceptance test most directly checks that failure mode?

  1. Query identical terms as HR and non-HR users
  2. Reindex a modified document and verify traceability
  3. Score the same question with controlled gold context
  4. Disable knowledge access and verify safe no-evidence behavior

Explanation

Correct Answer

A:

The scenario is that users have different rights to the same knowledge corpus; the defect to rule out is that agent cites a private HR document to unauthorized staff; testing via querying identical terms as HR and non-HR users challenges the design dependency where retrieval applies effective source entitlements to candidate documents.

Incorrect Answers

B:

Reindexing a modified document and verifying traceability verifies an adjacent objective where agents begin citing an obsolete article after an ingestion job; it leaves untested the causal path in which retrieval applies effective source entitlements to candidate documents for this case.

C:

This check exercises a different system behavior: separate relevance and groundedness tests isolate the failing layer; a pass there cannot establish that agent cites a private HR document to unauthorized staff no longer occurs in the required workflow.

D:

The suggested test serves policy answers must be tied to a current source, in which the agent confidently answers when retrieval fails; the present scenario instead concerns the reported problem where agent cites a private HR document to unauthorized staff.

Question 584

A release must prove that prompt changes fail to resolve a measured specialist error will not recur. The application is used in this situation: General models repeatedly misread rare engineering labels. Which acceptance test most directly checks that failure mode?

  1. Update an approved prompt and validate dependent flows in test
  2. Submit missing and malformed values and check validation behavior
  3. Install the agent to a pilot Microsoft 365 audience
  4. Evaluate held-out rare labels and previously unseen layouts

Explanation

Correct Answer

D:

The scenario is that general models repeatedly misread rare engineering labels; the defect to rule out is that prompt changes fail to resolve a measured specialist error; testing via evaluating held-out rare labels and previously unseen layouts challenges the design dependency where domain-trained models can be evaluated when generic baselines fall short.

Incorrect Answers

A:

In the alternate setting where multiple low-code apps need a common extraction prompt, the test exercises centralized reusable AI capabilities promote consistency and controlled updates; even a pass would not show that prompt changes fail to resolve a measured specialist error has been prevented.

B:

Submitting missing and malformed values and checking validation behavior verifies an adjacent objective where inspectors submit a short description and need suggested categorization; it leaves untested the causal path in which domain-trained models can be evaluated when generic baselines fall short for this case.

C:

Installing the agent to a pilot microsoft 365 audience verifies an adjacent objective where employees want curated assistance without owning a separate LLM runtime; it leaves untested the causal path in which domain-trained models can be evaluated when generic baselines fall short for this case.

Question 585

A policy review covers an agent for which real task mix differs from prelaunch performance tests. Its most relevant exposure is that router optimization could cut cost while materially worsening critical outcomes. What safeguard should be required before release?

  1. Set usage budgets, cost alerts and routing limits
  2. Budget for independent hosting, safety testing and monitoring
  3. Roll back route changes if quality guardrails regress
  4. Suppress assertions that require evidence when no source was retrieved
  5. Reject ambiguous identity matches rather than guessing

Explanation

Correct Answer

C:

The business setting is real task mix differs from prelaunch performance tests, with the exposure that router optimization could cut cost while materially worsening critical outcomes; the enforceable protection must correspond to the path where feedback-driven thresholds rebalance accuracy, latency and spend.

Incorrect Answers

A:

An agent scales from a trial to enterprise usage; its guardrail addresses the possibility that increased model traffic might run beyond the project budget without alerts; the present architecture must constrain the different threat that router optimization could cut cost while materially worsening critical outcomes.

B:

That protection is tied to the mechanism that owning the engine provides control over models, routing and execution; it does not prevent the distinct risk where router optimization could cut cost while materially worsening critical outcomes in the system under review.

D:

The alternate boundary mitigates a retrieval outage could make the model invent unsupported factual claims, which is relevant to required knowledge stores are intermittently unreachable; it leaves the current workflow vulnerable to router optimization could cut cost while materially worsening critical outcomes.

E:

That protection is tied to the mechanism that canonical keys and deduplication prevent incorrect cross-record joins; it does not prevent the distinct risk where router optimization could cut cost while materially worsening critical outcomes in the system under review.

Question 586

The architecture board has evidence that overall performance masks unequal escalation rates. The affected situation is: An AI triage model under-serves one customer language. What change most directly establishes the missing behavior?

  1. Evaluate outcomes for meaningful demographic or language disparities
  2. Implement harm-prevention testing before broader adoption
  3. Monitor newly released service agent behavior by customer cohort
  4. Regression-test case routing after service agent changes

Explanation

Correct Answer

A:

Without the intended behavior, overall performance masks unequal escalation rates; the proposed architecture establishes the specific design principle that segment-aware evaluations surface fairness risks for the process where an AI triage model under-serves one customer language.

Incorrect Answers

B:

This design would make sense if a helpful assistant can still produce unsafe financial or legal guidance, particularly where pilot satisfaction hides severe low-frequency failures; the requested architecture instead needs the dependency that segment-aware evaluations surface fairness risks because overall performance masks unequal escalation rates.

C:

For another type of architecture, cohort telemetry supports safe staged validation is useful; this design must instead support an AI triage model under-serves one customer language despite the problem where overall performance masks unequal escalation rates.

D:

In the case where a support agent changes how it classifies priority intents, the alternate design depends on controlled service evaluations validate routing before rollout; it does not correct the target failure, overall performance masks unequal escalation rates.

Question 587

A design adds autonomous behavior to a process where vendors send variable-layout invoices and shipping documents. This creates a documented risk: Uncertain tax identifiers extracted from invoices might be posted without review. Which enforced control is the best fit?

  1. Stop and compensate when a required approval fails
  2. Retain document version and ACL information on every chunk
  3. Validate types and authorization at the called service
  4. Review low-confidence financial fields before posting

Explanation

Correct Answer

D:

In the context where vendors send variable-layout invoices and shipping documents, the exposure is that uncertain tax identifiers extracted from invoices might be posted without review; containing it requires a guardrail in the operation where document processing models extract structured fields from document content.

Incorrect Answers

A:

Another process could need this guardrail because a flow might notify the customer even though a required approval failed; the design being assessed supports vendors send variable-layout invoices and shipping documents and has the explicit exposure that uncertain tax identifiers extracted from invoices might be posted without review.

B:

The safeguard for retrieved policy fragments lose context when documents are split deals with a different threat: chunked content might lose original permissions and version information; even after it is enforced, the exposure that uncertain tax identifiers extracted from invoices might be posted without review remains possible in the target workflow where document processing models extract structured fields from document content.

C:

The safeguard for the orchestrator chooses between similar invoice and credit tools deals with a different threat: an API might accept a valid-looking but unauthorized generated identifier; even after it is enforced, the exposure that uncertain tax identifiers extracted from invoices might be posted without review remains possible in the target workflow where document processing models extract structured fields from document content.

Question 588

Release reviewers will not accept a generic success message. The business scenario is that shared result caching makes answers fast across roles, and the historical failure is: One user's cached confidential answer appears for another user. Which test is appropriately targeted?

  1. Attempt a cross-environment write using staging credentials
  2. Disable knowledge access and verify safe no-evidence behavior
  3. Make identical queries as permitted and denied users
  4. Score the same question with controlled gold context

Explanation

Correct Answer

C:

Making identical queries as permitted and denied users is aligned with the failure that one user's cached confidential answer appears for another user; its causal relevance is that cache keys and access validation prevent security-context leakage for the situation where shared result caching makes answers fast across roles.

Incorrect Answers

A:

In the alternate setting where test automation must not reach live customer records, the test exercises environment-scoped identities isolate access and simplify incident containment; even a pass would not show that one user's cached confidential answer appears for another user has been prevented.

B:

This check exercises a different system behavior: evidence and refusal evaluations enforce source-supported claims; a pass there cannot establish that one user's cached confidential answer appears for another user no longer occurs in the required workflow.

D:

In a solution where an answer contradicts the handbook despite high model fluency, scoring the same question with controlled gold context could catch the defect that engineers tune prompts without checking whether evidence was retrieved; the current release gate must challenge the failure where one user's cached confidential answer appears for another user.

Question 589

An incident review has one confirmed symptom: New planning behavior occasionally leaks internal notes. For the use case where a hosted agent gains an additional research tool, which targeted diagnostic check would reveal the most pertinent evidence?

  1. Audit token use and secret inventory for agent services
  2. Diff tool traces and evaluation outcomes against baseline
  3. Audit send events and approver identities after rollout
  4. Compare approval records to agent risk classification

Explanation

Correct Answer

B:

In the setting where a hosted agent gains an additional research tool, diffing tool traces and evaluation outcomes against baseline can isolate the failure in which new planning behavior occasionally leaks internal notes, because evaluation suites validate quality and safety before publication.

Incorrect Answers

A:

Auditing token use and secret inventory for agent services explores a separate connection, where managed identity provides credential-less workload authentication where supported; the diagnosed requirement is a hosted agent gains an additional research tool, and the actual defect is new planning behavior occasionally leaks internal notes.

C:

The proposed inspection is associated with service agents draft customer messages but cannot send without review, where business approval checks remain independent of agent version; it would not expose the failure mode that new planning behavior occasionally leaks internal notes with the required specificity.

D:

Personal read-only assistants and payment agents have different risk; its matching diagnostic check is reasonable for every agent receives the same minimal review; the present investigation needs evidence for the different failure where new planning behavior occasionally leaks internal notes.

Question 590

An operational risk could arise because a shared MCP server might honor an unauthorized tool invocation. The AI system supports the business need that a Copilot Studio agent needs a reusable interface to governed operations. Which guardrail is designed for this threat?

  1. Restrict available actions and require confirmation for consequential changes
  2. Preserve table and row access controls for retrieval
  3. Narrow allowed tool capabilities and verify server-side authorization
  4. Enforce business-unit and row-level permissions at the data layer

Explanation

Correct Answer

C:

In the context where a Copilot Studio agent needs a reusable interface to governed operations, the exposure is that a shared MCP server might honor an unauthorized tool invocation; containing it requires a guardrail in the operation where MCP presents supported tools and resources through a common integration contract.

Incorrect Answers

A:

This restriction is associated with employees express several intents and need interactive clarification and addresses the exposure that a conversational tool could invoke consequential changes from an ambiguous request; it does not specifically protect the target workflow against a shared MCP server might honor an unauthorized tool invocation.

B:

The proposed control targets Dataverse retrieval might reveal a row not authorized to the caller, while the given risk scenario is a shared MCP server might honor an unauthorized tool invocation; these exposures arise in different operational contexts.

D:

The alternate boundary mitigates shared data tables might expose rows to a different business unit, which is relevant to agents and apps need shared definitions of customers and work orders; it leaves the current workflow vulnerable to a shared MCP server might honor an unauthorized tool invocation.

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