McAfee-Secured Website

Exam Code: AI-103

Exam Name: Developing AI Apps and Agents on Azure

Certification Provider: Microsoft

Corresponding Certification: Microsoft Certified: Azure AI Apps and Agents Developer Associate

100% Microsoft AI-103 Guaranteed Success With Testking Training Materials

Get Real AI-103 Exam Questions & Expert Verified Answers as Experienced on the Real Exam!

187 Questions & Answers with Testing Engine

"Developing AI Apps and Agents on Azure Exam", also known as AI-103 exam, is a Microsoft certification exam.

Pass your tests with the always up-to-date AI-103 Exam Engine. Your AI-103 training materials keep you at the head of the pack!

Satisfaction Guaranteed

Money Back Guarantee

Testking provides hassle-free money back guarantee with our products. That is because we have 100% trust in the abilities of our professional and experience product team, and our record is a proof of that.

99.6% PASS RATE
Was: $137.49
Now: $124.99

Microsoft AI-103 Practice Test 15

Question 140

An analyst bot creates findings from several uploaded compliance reports. Auditors must trace each conclusion to a specific revision of source evidence. What should be recorded?

  1. Record provenance for generated findings linked to source documents.
  2. Keep only the answer text and the model's subjective confidence rating.
  3. Infer citation page numbers from retrieval rank instead of the actual source-document location.
  4. Replace original documents with a generated narrative after processing.

Explanation

Correct Answer

A:

Provenance metadata makes generated claims inspectable against the actual source version.

Incorrect Answers

B:

Confidence without source links cannot establish which evidence supports a finding.

C:

Retrieval position is not a trustworthy page locator or provenance record.

D:

Destroying the source removes the audit trail needed for independent validation.

Question 141

A customer-data incident requires tracing a suspect agent request across tools without storing raw secrets. Which evidence should be retained?

  1. Generate trace IDs after the incident by asking the model to invent representative values.
  2. Retain trace IDs to support incident investigation without exposing secrets.
  3. Delete all failed tool-call events because they might contain private information.
  4. Save every authorization token in plain text for future incident replay.

Explanation

Correct Answer

B:

Correlated nonsecret metadata supports reconstruction while reducing the exposure of payload credentials.

Incorrect Answers

A:

Fabricated identifiers cannot reconstruct real service activity.

C:

Removing failures erases information essential to investigating unauthorized attempts.

D:

Raw tokens may still be valid and can create additional compromise paths.

Question 142

A new public chatbot version appears more helpful but occasionally bypasses a safety refusal. What quality evidence should gate release?

  1. Run representative allowed and adversarial safety evaluations, require the defined acceptance thresholds before promotion.
  2. Approve the version once it passes an HTML color-contrast check.
  3. Disable content filters during the test to avoid lowering response rate.
  4. Release based solely on average user satisfaction from friendly prompts.

Explanation

Correct Answer

A:

Versioned safety tests catch regressions and can balance false positives and prohibited outputs.

Incorrect Answers

B:

Visual accessibility testing does not verify moderation safeguards.

C:

A test without production safety controls does not measure the release behavior.

D:

Benign satisfaction scores do not test disallowed or manipulative inputs.

Question 143

A model upgrade starts blocking benign medical training diagrams. Safety incidents did not rise, but legitimate users are rejected. What metric matters?

  1. Remove all safety checks rather than analyzing permitted cases.
  2. Treat every block as proof the model prevented actual harm.
  3. Measure only the size of generated image files.
  4. Measure false-positive moderation effects on legitimate business requests.

Explanation

Correct Answer

D:

False positives quantify legitimate content incorrectly blocked and help calibrate interventions.

Incorrect Answers

A:

Uncontrolled disabling can expose users to genuinely prohibited outputs.

B:

Blocked benign content is not a confirmed policy violation.

C:

Image bytes do not indicate moderation classification accuracy.

Question 144

An insurance assistant cites three policy clauses but adds an uncited explanation of what regulators would likely allow. What response discipline is needed?

  1. Increase sampling temperature and interpret fluency as legal certainty.
  2. Require model outputs to distinguish retrieved fact from generated synthesis.
  3. Combine all model reasoning with source quotations without labeling the boundary.
  4. Delete citations to make the narrative seem more coherent.

Explanation

Correct Answer

B:

Distinguishing extracted evidence from synthesis prevents speculation being presented as confirmed policy.

Incorrect Answers

A:

Creative language is not evidence of regulatory authorization.

C:

Users cannot tell which claims are supported when evidence and conjecture are mixed.

D:

Removing sources weakens traceability of regulatory claims.

Question 145

An employee asks a policy question for which the authorized search index returns no matching passages. What is the safe response pattern?

  1. Invent the most likely policy and present it as an official answer.
  2. Output a made-up citation to make the answer look evidence-based.
  3. Add user-facing uncertainty when retrieval returns insufficient evidence.
  4. Switch to an unauthorized index to obtain any text that sounds relevant.

Explanation

Correct Answer

C:

An absence-aware response avoids fabricating organizational rules when retrieval lacks support.

Incorrect Answers

A:

Plausibility is not proof of an actual internal policy.

B:

False provenance disguises rather than fixes absent evidence.

D:

Source access restrictions still apply when the authorized corpus lacks an answer.

Question 146

A service agent generated work order WO-318 after a user request. Internal audit requires proof of the approved customer ticket that justified it. What linkage should exist?

  1. Log agent actions with validated ticket and approval references.
  2. Omit all tool-result IDs to keep the logs small.
  3. Log the latest ticket ID in the system rather than the ticket actually used in the agent action.
  4. Save only the model's final message that it completed the order.

Explanation

Correct Answer

A:

An action-to-authorization trail supports review of whether the order was created for a legitimate case.

Incorrect Answers

B:

Without outcome IDs, auditors cannot tie the agent decision to the real work order.

C:

A convenient but mismatched ticket identifier fails the audit requirement to bind an action to its approved case.

D:

Generated wording is not sufficient to prove the transaction or its authorization.

Question 147

An agent can check the current warehouse inventory and propose a reorder but may not move stock or delete records. Which execution policy implements the scope?

  1. Allow delete operations when an inventory API is temporarily unreachable.
  2. Expose every ERP operation and ask the model not to use dangerous ones.
  3. Let the user choose any hidden tool by describing it in a prompt.
  4. Limit autonomous workflow scope to an allowlisted set of tools.

Explanation

Correct Answer

D:

Explicit tool selection and scoped credentials prevent unapproved mutation even if model text requests it.

Incorrect Answers

A:

Backend failure does not expand the agent's authorized business scope.

B:

Text restrictions cannot reliably compensate for overprivileged tools.

C:

Untrusted requests should not determine an unrestricted execution surface.

Question 148

A helpdesk agent can reset user accounts, but irreversible deprovisioning requires a security officer. What should guard a delete command?

  1. Make all users security officers to simplify the tool permission model.
  2. Treat the agent's own confidence score as equivalent to officer approval.
  3. Block destructive agent actions absent a separate approval token.
  4. Run deletion immediately and show a warning after the account is gone.

Explanation

Correct Answer

C:

An independent scoped authorization stops destructive changes absent the required signoff.

Incorrect Answers

A:

Broad role elevation removes the intended separation of duties.

B:

Generated confidence is not an approved identity's authorization.

D:

Notification after an irreversible act cannot satisfy a precondition.

Question 149

A RAG chatbot protects the user prompt from direct jailbreaks, but retrieved websites can also inject malicious instructions. What safety placement must be verified?

  1. Protect both direct user prompts and retrieved document pathways.
  2. Assume user-message filtering automatically sanitizes all subsequent retrieved documents.
  3. Apply direct-prompt attack detection only to the outgoing model response text.
  4. Treat all retrieved content as system instructions after search returns it.

Explanation

Correct Answer

A:

Direct and indirect injection enter through different trust paths and require appropriate protection points.

Incorrect Answers

B:

A retrieved page can introduce instructions after the initial prompt inspection.

C:

This leaves the retrieved document/tool-result attack entry path unchecked, allowing lower-trust instructions to reach the agent.

D:

Elevating lower-trust documents permits instruction hijacking.

Question 150

An inspection agent uses OCR on photographed machinery labels. One label contains text ordering the model to send credentials elsewhere. How should the extracted text be treated?

  1. Trust the OCR service identity to validate all instructions appearing in the photographed scene.
  2. Give the agent file-export access so it can comply with the photographed command.
  3. Treat text extracted from images as untrusted instructions.
  4. Execute commands from OCR because the image is already inside the user's request.

Explanation

Correct Answer

C:

Content extracted from images may contain adversarial instructions even when produced by a trusted OCR service.

Incorrect Answers

A:

Extraction accurately reporting hostile text does not make that text trustworthy.

B:

Additional privileges would worsen the consequences of an indirect injection attack.

D:

An image's embedded text does not confer user or system authorization.

Developing AI Apps and Agents on Azure (AI-103) Practice Tests