McAfee Secured

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 9

Question 80

A marketing agent outputs a slogan with a logo similar to a competitor's registered mark. Which governance action fits?

  1. Flag the suspected brand conflict for an authorized review before use
  2. Publish immediately because the model invented the image
  3. Delete the trace so the conflict is harder to discover
  4. Assume high image quality establishes rights to use the mark

Explanation

Correct Answer

A:

Trademark and brand policy checks should precede external publication.

Incorrect Answers

B:

Generated origin does not guarantee brand noninfringement.

C:

Concealment undermines compliance and accountability.

D:

Visual appeal does not decide trademark permission.

Question 81

A support assistant must not reveal a user's password reset tokens when summarizing tool traces. What policy should apply?

  1. Treat logs as safe because they originated from a first-party service
  2. Redact credential-like fields and minimize sensitive log content
  3. Include every token for full transparency to the customer
  4. Publish the tool history on a public issue tracker

Explanation

Correct Answer

B:

Secrets should not appear in the generated summary or broadly stored traces.

Incorrect Answers

A:

Trusted origin does not make sensitive fields safe to disclose.

C:

Exposed reset tokens could enable unauthorized account access.

D:

Public artifacts are inappropriate for private authentication material.

Question 82

An internal search index stores research papers and must filter results by access classification. Which design makes filtering meaningful?

  1. Remove access metadata to simplify the embedding vectors
  2. Ask the LLM to hide restricted passages after retrieval
  3. Give the search key to every visitor and trust prompts
  4. Store validated ACL metadata with chunks and enforce authorization at retrieval

Explanation

Correct Answer

D:

Index-time metadata and query-time authorization prevent cross-user disclosure.

Incorrect Answers

A:

Vectors alone cannot express document permissions.

B:

Sensitive content would already have entered an unauthorized model context.

C:

Shared unrestricted keys bypass scoped authorization.

Question 83

A user needs to locate a spoken phrase inside a recorded training video. What retrieval metadata is essential?

  1. Only the video's initial thumbnail
  2. Only the number of frames rendered
  3. Transcript text with segment timestamps and original video reference
  4. Only generated replacement subtitles with no source alignment

Explanation

Correct Answer

C:

Time-aligned search links recognized words to the playback location.

Incorrect Answers

A:

A still image does not show where speech occurred.

B:

Frame count cannot identify spoken phrases.

D:

Synthetic text may not match the recording.

Question 84

A retrieval app indexes diagrams but returns no matches for concepts shown only as arrows and blocks. What can improve coverage?

  1. Generate grounded figure descriptions or image embeddings during ingestion
  2. Discard figures when OCR finds little text
  3. Invent labels for every diagram without preserving source
  4. Index only surrounding paragraph lengths

Explanation

Correct Answer

A:

Visual semantics must be represented in the searchable index.

Incorrect Answers

B:

Diagrams may convey meaning without many printed words.

C:

Ungrounded additions can contaminate search results.

D:

Text-length metadata cannot describe connections inside diagrams.

Question 85

A customer support agent can issue refunds up to $50 automatically but needs approval above that amount. Where should the threshold be enforced?

  1. At the refund tool boundary using amount and verified authorization
  2. Only by hiding the refund tool description
  3. Only in a chat message posted after the refund
  4. Only by lowering model temperature

Explanation

Correct Answer

A:

An application invariant protects money-moving operations despite prompt drift.

Incorrect Answers

B:

Undocumented capabilities remain callable if they are exposed.

C:

A retrospective warning cannot undo an unauthorized transfer.

D:

More deterministic text generation does not enforce spend limits.

Question 86

A billing assistant must produce machine-readable line items, and the model returns amounts as strings containing currency symbols. What should be enforced?

  1. Typed numeric fields with currency explicitly represented and validated
  2. Allow every amount as arbitrary unparsed prose
  3. Convert all amounts to integers by dropping decimals
  4. Use the model's final paragraph as the only parser input

Explanation

Correct Answer

A:

Structured values separate amount semantics from display formatting.

Incorrect Answers

B:

Downstream arithmetic becomes unreliable.

C:

Truncation can alter financial totals.

D:

Free-text parsing is less robust than a defined output schema.

Question 87

A document assistant retrieves a private file but the querying user should not access it. Where must access enforcement occur?

  1. Only with a low model temperature
  2. Before document content enters the model, using permission-filtered retrieval
  3. Only by encrypting the final answer after unauthorized retrieval
  4. Only in the final prompt asking the assistant to hide it

Explanation

Correct Answer

B:

The source cannot be safely unlearned after an unauthorized passage is supplied.

Incorrect Answers

A:

Determinism does not enforce source permissions.

C:

Output encryption does not repair data access policy violation.

D:

The material already crossed the user's access boundary.

Question 88

A RAG product shows worse grounded answers after a search reindex. Which TWO diagnostics matter most?

  1. Query relevance tests with known source-claim matches
  2. Only the number of active employees
  3. Indexer completeness, source freshness and rejected-document counts
  4. Only the model's marketing release date
  5. Only change the model generation settings while keeping the broken retrieval index

Explanation

Correct Answers

A:

Retrieval evaluation identifies whether relevant passages remain findable.

C:

Lost or stale indexed evidence can cause unsupported model responses.

Incorrect Answers

B:

User count does not measure grounded retrieval correctness.

D:

The model may be unchanged while its search corpus drifted.

E:

Sampling adjustments cannot restore missing documents or correct retrieval relevance.

Question 89

A planning team needs lightweight sentiment labeling across millions of short reviews. What should it compare before choosing a large generative model?

  1. Only maximum output tokens as the selection metric
  2. A narrowly capable classifier or small model against an evaluated sample
  3. Automatic multi-agent debate on every two-word review
  4. A text-to-video model to infer reviewer intent

Explanation

Correct Answer

B:

A targeted model may deliver required labels at lower operational cost.

Incorrect Answers

A:

Token allowance does not establish classification accuracy.

C:

Several models add needless latency to a simple classification decision.

D:

Video synthesis does not perform the required text sentiment task.

Question 90

A scanned ID document includes a machine-readable zone and printed name. What quality check helps?

  1. Compare recognized fields across reliable regions and flag inconsistencies
  2. Trust any single recognized line without comparison
  3. Use OCR on the printed name without validating the machine-readable zone
  4. Replace unclear characters with random plausible letters

Explanation

Correct Answer

A:

Cross-field validation can catch OCR errors before identity-sensitive use.

Incorrect Answers

B:

One OCR pass may misread critical characters.

C:

An isolated OCR result may contain errors that cross-region validation could reveal.

D:

Fabrication corrupts official identity data.

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