Microsoft AB-100 Practice Test 60

Question 590

An airline customer moves from a flight-change topic to a refund topic. The previous topic already submitted a rebooking request, but its response was delayed. How should the handoff prevent duplicate changes?

Start a fresh rebooking action each time the dialog enters a new topic.
Discard the booking identifier as soon as the refund topic begins.
Return “rebooking complete” because the previous topic generated a confident acknowledgement.
Carry a stable booking correlation ID and query the rebooking operation state before attempting another mutation.

Explanation

Correct Answer

Topic transitions do not erase outstanding transaction state; reconciliation avoids two rebookings for one intent.

Incorrect Answers

New action IDs prevent deduplication and can create multiple purchases or reservations.

Losing the original transaction reference makes safe compensation and status checks harder.

Conversational success text does not prove the external booking service committed the operation.

Question 591

A sales team introduces a “renew contract” topic. After promotion, “renew quote” requests are incorrectly routed to contract renewal. What is the most targeted release correction?

Refine overlapping triggers/entities and regression-test quote-renewal and contract-renewal utterances against expected routes.
Disable all generative answers throughout the tenant.
Give contract renewal priority for every phrase containing “renew.”
Automatically invoke both renewal APIs and keep the one that returns faster.

Explanation

Correct Answer

Intent-routing acceptance should cover near-neighbor phrases where the wrong action has significant business consequences.

Incorrect Answers

Eliminating broad generation does not solve overlap between two specialized business topics.

A broad keyword priority reproduces the misrouting of quote-specific requests.

The APIs can perform different irreversible operations and response latency does not establish user intent.

Question 592

A contract-help assistant retrieves two clauses with identical titles: one expired yesterday and one effective today. The semantic score favors the expired clause. Which retrieval pipeline step is decisive?

Use both clauses as equally binding without resolving their conflicting validity dates.
Filter by approved effective interval and source version before ranking passages for generation.
Select the passage with the longest body because it contains more details.
Choose the first indexed result and omit its revision identifier from the answer.

Explanation

Correct Answer

Version authority and time validity must constrain which clauses are even eligible as grounding evidence.

Incorrect Answers

Merging superseded and active terms can create a policy statement that neither source authorizes.

Document length does not prove which agreement revision controls the present request.

Index order can select obsolete policy and conceals the provenance needed for review.

Question 593

A support portal indexes product repair notes and private customer incident reports together. A dealer can access generic repair notes but not customer-specific records. Where must the access rule apply?

In a system prompt that asks the assistant not to mention private incidents.
After a complete answer is generated, by redacting customer names only.
At the retrieval service before any customer incident excerpt becomes model context.
Only in CSS that hides customer names on the final chat page.

Explanation

Correct Answer

The authorization boundary must trim results to dealer entitlements, including cached and re-ranked passages.

Incorrect Answers

Instruction text does not stop privileged search from returning restricted source content.

Post-generation redaction cannot reliably remove sensitive facts inferred from unauthorized records.

Presentation hiding cannot remove confidential material already delivered to the model or tracing service.

Question 594

A compliance copilot cites `Policy Handbook` for an overtime answer, but the search connector provides no passage locator or revision ID. Which remediation supports an audit-ready citation?

Use the model’s footnote numbering without storing the underlying retrieval reference.
Add the name of the latest search query as the document citation.
Hide citation links and trust the model to recall them later.
Attach source document identity, approved revision, and passage position to each retrieved chunk and verify links resolve.

Explanation

Correct Answer

Citations must identify the actual controlling evidence, not merely an attractive document title.

Incorrect Answers

Numbered footnotes are not proof of provenance when no source chunk can be reproduced.

A query string identifies a request, not which approved paragraph supported the conclusion.

Unreproducible references do not meet compliance traceability and can mask a stale source.

Question 595

A procurement agent indexes scanned rate tables with column headings on one page and prices on the next. OCR text flattening causes rates to attach to the wrong service level. What should preprocessing preserve?

Table structure, headers, row associations, page references and confidence flags before embedding the content.
Remove every header so the embedding model focuses on numeric amounts.
Infer missing row identities from whichever line appears nearest in the raw OCR stream.
Ask the model to choose a price and then search for a matching service name.

Explanation

Correct Answer

Correct column-to-price relationships are necessary for a grounded price comparison; plain text order may corrupt them.

Incorrect Answers

Without headers the meaning, units, and effective service tiers of each price are ambiguous.

Physical text proximity cannot establish multi-page table relationships without structural validation.

Back-fitting a source after answering can produce fabricated associations instead of evidence-based retrieval.

Question 596

A claims chatbot uses a shared embedding-response cache. After a user from insurer A asks a question, insurer B receives a similar answer containing A’s claim details. Which cache design is needed?

Raise the cache similarity threshold while retaining a shared privileged cache.
Scope cache keys and retrieval permissions by authenticated tenant and relevant entitlement version.
Keep one global semantic cache and redact only insurer names from responses.
Tell the answer generator never to reveal facts from another user.

Explanation

Correct Answer

Semantically similar prompts cannot share restricted records across tenant identities or revoked permissions.

Incorrect Answers

Even nearly identical requests may refer to different confidential claims and require tenant isolation.

Claim facts remain sensitive even if a company name is hidden, and global reuse bypasses authorization.

Prompts cannot enforce cache partitioning when data is already provided to the generator.

Question 597

A service representative has a Canvas app showing an authorized incident record and wants a “Summarize case” button. The agent must not gain access to other customers. What integration should be used?

Use the visible customer name as the only identity key.
Grant the app’s agent environment-administrator privileges to eliminate authorization prompts.
Pass a stable record ID to an authorized agent action and enforce row-level scope in the service before returning the summary.
Send the full Dataverse table to the model and ask it to use the selected row only.

Explanation

Correct Answer

An app selection is not a substitute for verifying row access at the agent’s data-access boundary.

Incorrect Answers

Duplicate display names can misidentify records and enable cross-customer disclosures.

An administrative identity breaks least privilege for a read-only case-summary task.

Unnecessary broad data disclosure defeats record-level access and increases leak risk.

Question 598

A Canvas app calls an AI assistant to draft a supplier change. Before it can submit an approved supplier-bank update, which TWO independent controls are required? (Select TWO.)

Reuse one shared Power Apps administrator token for all users.
Validate the supplier record and requesting user’s permission in a server-side action.
Require the configured independent finance approver to authorize the bank-detail change before commit.
Let the language model certify that the caller appears trustworthy.
Hide the Submit button after the model has already called the write API.

Explanation

Correct Answers

Server-side authorization binds the request to the correct supplier and caller, regardless of prompt text.

Bank-detail updates are high impact and need an enforceable approval event rather than narrative consent.

Incorrect Answers

A global token removes individual accountability and access scoping across suppliers.

Model confidence cannot replace a verifiable account identity and role check.

UI visibility cannot withdraw an action that has already crossed the transaction boundary.

Question 599

A field inspector’s Canvas app uses an agent to create maintenance tickets. The tablet is offline and the cloud action cannot confirm ticket creation. What should the UI report?

Show a pending local draft and retry under a stable idempotency key once connectivity returns, without claiming creation.
Show “ticket submitted” immediately because the model produced a well-formed payload.
Generate a different ticket number every time the offline workflow retries.
Discard the inspector’s draft when the request times out.

Explanation

Correct Answer

Until the service returns a durable ticket ID, the offline action has not been verified as committed.

Incorrect Answers

Payload syntax does not prove that a remote ticket service accepted or stored the request.

Unique retry numbers can create duplicate incidents rather than reconcile the same intended write.

Losing the draft destroys evidence needed for controlled recovery and resubmission.

Question 600

A Canvas app stores a private escalation note in a client-visible variable and passes it to an agent, although the user is not entitled to see the note. What architectural correction is required?

Hide the note control using a conditional Visible property only.
Fetch authorized escalation context server-side and return only fields permitted for the current app user.
Encode the note with Base64 before storing it in the app variable.
Ask the agent to ignore restricted notes if the user requests them.

Explanation

Correct Answer

Client variables can be inspected and are not a secure data-access boundary for restricted notes.

Incorrect Answers

A hidden control does not establish server-side permission filtering or prevent data extraction.

Encoding does not provide authorization or confidentiality to someone who can read client state.

The restricted content should not have reached the user-accessible app context in the first place.

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

Microsoft AB-100 Practice Test 58 - 10 Questions (Last Reviewed 13-Nov-2026)
Microsoft AB-100 Practice Test 59 - 10 Questions (Last Reviewed 12-Nov-2026)
Microsoft AB-100 Practice Test 60 - 10 Questions (Last Reviewed 11-Nov-2026)
Microsoft AB-100 Practice Test 61 - 10 Questions (Last Reviewed 10-Nov-2026)
Microsoft AB-100 Practice Test 62 - 10 Questions (Last Reviewed 09-Nov-2026)

Pass Your Microsoft AB-100 Exam Easy!

Question 590

Explanation

Correct Answer

Incorrect Answers

Question 591

Explanation

Correct Answer

Incorrect Answers

Question 592

Explanation

Correct Answer

Incorrect Answers

Question 593

Explanation

Correct Answer

Incorrect Answers

Question 594

Explanation

Correct Answer

Incorrect Answers

Question 595

Explanation

Correct Answer

Incorrect Answers

Question 596

Explanation

Correct Answer

Incorrect Answers

Question 597

Explanation

Correct Answer

Incorrect Answers

Question 598

Explanation

Correct Answers

Incorrect Answers

Question 599

Explanation

Correct Answer

Incorrect Answers

Question 600

Explanation

Correct Answer

Incorrect Answers

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