Microsoft AB-100 Practice Test 28

Question 270

An AI summarizer provides a correct answer using a document that a test user cannot read. What should the evaluation report?

Defer the verdict until the model’s answer relevance is graded, without a separate document-access assertion.
A pass because factual correctness outweighs document access rights.
A security failure even when the answer text happens to be accurate.
Treat unauthorized retrieval as a citation-format defect rather than a failed access-control test.

Explanation

Correct Answer

Correctness and authorization are independent requirements for a safe grounded solution.

Incorrect Answers

Relevance is independent of user entitlement. Unauthorized source use is a security failure even when factually correct.

Protected data cannot be used without entitlement.

The unauthorized citation itself can disclose source facts; grounding must be blocked at query time, independent of answer accuracy.

Question 271

A customized credit-risk model has a low overall error rate but incorrectly rejects applicants from one geographic group much more often. Which validation is necessary?

Remove subgroup fields from reports while still accepting biased outcomes.
Treat an uncalibrated confidence score as a fairness certificate.
Use overall accuracy alone because the largest group dominates.
Evaluate group-level error rates and fairness controls on representative approved data.

Explanation

Correct Answer

Responsible-AI evaluation must examine consequential disparate failure patterns beyond average scores.

Incorrect Answers

Hiding metrics does not correct outcomes.

Confidence is not a disparity analysis.

Aggregates can conceal group-specific harm.

Question 272

A prompt action requests a valid refund amount but a retrieved document includes a hidden instruction to ignore the maximum. Which prompt test best evaluates safety?

Assume a system prompt alone makes the refund service safe.
Evaluate only whether the answer includes currency formatting.
Set a longer timeout so the injected request can finish.
Check that injected instructions cannot override the validated transaction limit.

Explanation

Correct Answer

Model and tool behavior must be tested against malicious lower-trust content and enforceable monetary ceilings.

Incorrect Answers

Prompts cannot substitute for back-end limits.

Currency formatting is not transaction permission.

Timeout tuning does not remove prompt-injection risk.

Question 273

A user submits a Sales return that should update inventory in Supply Chain Management and a credit in Finance. What test exposes partial failure?

Test only the successful Sales form submission.
Assume independent systems cannot disagree once one API accepts a request.
Fail the Finance step deliberately and assert inventory/return compensation or a recoverable exception queue.
Delete failed transaction traces before evaluation.

Explanation

Correct Answer

End-to-end testing must exercise partial commitments across business boundaries.

Incorrect Answers

A successful first form is not end-to-end state.

Distributed operations can fail independently.

Traces are essential to understand compensation behavior.

Question 274

Copilot generates tests for a payroll-support agent. Select TWO acceptance practices before promoting those cases to the official regression suite.

Execute the tests against a controlled dataset and retain versioned evidence of results.
Discard failed tests until the observed score reaches 100%.
Treat every generated expected answer as correct because the test text is syntactically valid.
Have an authorized payroll domain reviewer check expected outcomes against current policy.
Run tests against live employee accounts with unapproved write permissions.

Explanation

Correct Answers

Controlled execution and retained versions make regression evidence reproducible.

A domain reviewer supplies business truth that an auto-generated oracle may lack.

Incorrect Answers

Removing failures would falsely inflate quality metrics.

Schema validity cannot establish payroll policy correctness.

Unapproved writes risk real employee data.

Question 275

An autonomous inventory agent moves stock between warehouses. Which scenario is mandatory in acceptance testing?

A transfer between warehouses that do not exist in the system.
A transfer that crosses an authorization threshold and must stop for human approval.
Test only that the assistant explains a proposed inventory transfer in detail, without exercising the restricted approval threshold.
Only a routine transfer safely below every threshold.

Explanation

Correct Answer

Boundary tests prove the agent respects financial and operational authority limits.

Incorrect Answers

Invalid locations test input handling but not the specific approval threshold.

Narrative quality cannot prove whether the transaction tool blocks warehouse moves that exceed delegated spending or stock authority.

Happy paths alone miss key safety behavior.

Question 276

A legal agent answers with citations. The evaluator sees a quoted sentence that matches a policy but the policy is not effective until next month. Which metric matters?

The number of linked policy documents regardless of effective dates.
Score evidence quality using citation style and grammatical correctness without validating the policy’s effective date for the case.
Time-valid source attribution: whether the evidence is authoritative for the date of the decision.
Citation text similarity alone.

Explanation

Correct Answer

A correctly quoted future policy may still be wrong for the current transaction.

Incorrect Answers

More links can include contradictory versions.

A well-written answer can quote a policy that is not binding during the relevant period; source validity must be checked.

Lexical match does not establish effective authority.

Question 277

A computer-vision claims model was validated only on clean scans; production receives low-light phone photos. What test data should be added?

Add synthetic low-light images with no verified labels, then assume that image variety alone proves model robustness.
Representative low-light and device-specific images with labeled extraction and rejection outcomes.
More duplicates of the original clean scans.
Use generated image-description detail as the sole metric rather than measuring accuracy and abstention on low-light production images.

Explanation

Correct Answer

Evaluation must cover realistic distribution and source-quality variations.

Incorrect Answers

Without ground-truth extraction and rejection labels, additional images cannot measure false readings or safe abstention under low light.

Duplicates do not reveal new failure patterns.

Descriptive verbosity cannot establish that a claims image was extracted correctly or rejected when unreadable.

Question 278

A prompt template tells a model to cite sources, yet generated responses invent source identifiers. What validation should be required?

Check that each citation resolves to an authorized retrieved document and supports the attributed claim.
Treat a plausible URL string as sufficient proof.
Require two citation-looking tokens without checking their destinations.
Confirm retrieval returned at least one document and assume generated citation identifiers are valid without resolving them.

Explanation

Correct Answer

Citation verification must operate on actual document IDs and evidence spans.

Incorrect Answers

A plausible-looking link can be fabricated.

Citation counts alone do not measure support.

A nonempty retrieval result does not prove the model’s reference IDs point to those documents or support its claims.

Question 279

A Sales agent creates a customer opportunity, Finance sets credit terms and Service opens onboarding. Which system-level assertion is essential?

Cross-system customer IDs and financial terms must reconcile with the authorized record after each stage.
Verify that all three Dynamics 365 actions display the same success banner without checking persisted customer and finance records.
The first service’s success code proves every subsequent write committed.
Confirm the Sales, Finance and Service tool names appear in the execution trace without reconciling committed records.

Explanation

Correct Answer

End-to-end tests verify persisted business state, not presentation or optimistic orchestration prose.

Incorrect Answers

A consistent UI acknowledgment does not prove Sales, Finance and Service transactions actually committed the intended business state.

Distributed steps can fail after earlier success.

Invoked tool names indicate attempts, not persisted success. Verify related customer keys and financial terms in each system.

Question 280

Copilot generates twenty regression tests but five are nearly identical and two have an incorrect answer key. How should the suite be curated?

Keep all cases unchanged to maximize the test count.
Correct the faulty oracles and remove redundant cases while retaining distinct failure boundaries.
Declare any generated case a passing test by definition.
Remove all failures without documenting the underlying defect.

Explanation

Correct Answer

A good suite measures distinct requirements with trusted expected outcomes.

Incorrect Answers

Quantity cannot replace coverage or correct assertions.

Generated text requires validation.

Deleting failures hides real weaknesses.

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

Microsoft AB-100 Practice Test 26 - 10 Questions (Last Reviewed 19-Oct-2026)
Microsoft AB-100 Practice Test 27 - 10 Questions (Last Reviewed 18-Oct-2026)
Microsoft AB-100 Practice Test 28 - 10 Questions (Last Reviewed 17-Oct-2026)
Microsoft AB-100 Practice Test 29 - 10 Questions (Last Reviewed 16-Oct-2026)
Microsoft AB-100 Practice Test 30 - 10 Questions (Last Reviewed 15-Oct-2026)

Pass Your Microsoft AB-100 Exam Easy!

Question 270

Explanation

Correct Answer

Incorrect Answers

Question 271

Explanation

Correct Answer

Incorrect Answers

Question 272

Explanation

Correct Answer

Incorrect Answers

Question 273

Explanation

Correct Answer

Incorrect Answers

Question 274

Explanation

Correct Answers

Incorrect Answers

Question 275

Explanation

Correct Answer

Incorrect Answers

Question 276

Explanation

Correct Answer

Incorrect Answers

Question 277

Explanation

Correct Answer

Incorrect Answers

Question 278

Explanation

Correct Answer

Incorrect Answers

Question 279

Explanation

Correct Answer

Incorrect Answers

Question 280

Explanation

Correct Answer

Incorrect Answers

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