McAfee Secure

Exam Code: AI-901

Exam Name: Microsoft Azure AI Fundamentals

Certification Provider: Microsoft

Corresponding Certification: Microsoft Certified: Azure AI Fundamentals

Microsoft AI-901 Questions & Answers

Study with Up-To-Date REAL Exam Questions and Answers from the ACTUAL Test

166 Questions & Answers with Testing Engine

"Microsoft Azure AI Fundamentals Exam", also known as AI-901 exam, is a Microsoft certification exam.

Pass your tests with the always up-to-date AI-901 Exam Engine. Your AI-901 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-901 Practice Test 52

Question 510

An Azure-based AI team must satisfy this objective: A chatbot provides outdated pricing because its prompt includes last month's catalog. Which design choice is best?

  1. Define required output sections and link citations to actual retrieved passages
  2. Require explicit confirmation and backend authorization before tool execution
  3. Retrieve current approved pricing evidence instead of embedding volatile facts in instructions
  4. Pass relevant conversation context or a verified summarized state within token limits

Explanation

Correct Answer

C:

Static prompts are unsuitable as a changing source of commercial truth. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

Explicit response structure makes evidence easier to verify. That is not sufficient here: Static prompts are unsuitable as a changing source of commercial truth.

B:

High-impact side effects need permissions independent of conversational language. That is not sufficient here: Static prompts are unsuitable as a changing source of commercial truth.

D:

Stateful conversations require supplying or retrieving relevant prior context. That is not sufficient here: Static prompts are unsuitable as a changing source of commercial truth.

Question 511

An AI operations team is testing two distinct capabilities. Failure (1): The system always assumes the loudest phrase is the request. Failure (2): A user can change inventory by entering instructions into a prompt only. Which TWO specific corrections should the team apply?

  1. Run unsupervised clustering and evaluate cluster characteristics
  2. Use layout-aware structured invoice extraction
  3. Train or configure intent classification with a bounded label set
  4. Evaluate noise conditions and request clarification at low confidence
  5. Enforce backend identity and method scopes

Explanation

Correct Answers

D:

Correct for failure (1). Speech input quality affects subsequent language classification. The proposed correction directly addresses that reported defect.

E:

Correct for failure (2). Agentic workflows combine information access with controlled actions. It independently corrects the second defect.

Incorrect Answers

A:

Clustering discovers groups based on feature similarity. It does not complete the two decisions required in this question.

B:

Table extraction associates quantities, descriptions and prices by row. The stated pair of failures instead requires “Evaluate noise conditions and request clarification at low confidence” and “Enforce backend identity and method scopes”; this option cannot substitute for both.

C:

Classification maps text to one of several predefined categories. The stated pair of failures instead requires “Evaluate noise conditions and request clarification at low confidence” and “Enforce backend identity and method scopes”; this option cannot substitute for both.

Question 512

An AI operations team is testing two distinct capabilities. Failure (1): Parsing drops headings and reading order from tagged PDFs. Failure (2): The system adds fabricated citations when search returns no documents. Which TWO specific corrections should the team apply?

  1. Gate the payment tool with verified role and amount checks
  2. Require evidence-backed citation construction and abstention on no results
  3. Run native-language acceptance cases with verified translations and experts
  4. Add post-retrieval and pre-action policy enforcement
  5. Use layout-preserving extraction and test alternatives

Explanation

Correct Answers

B:

Correct for failure (2). Honesty requires acknowledging when a claim cannot be substantiated. It independently corrects the second defect.

E:

Correct for failure (1). Inclusion requires robust input handling beyond a single document template. The proposed correction directly addresses that reported defect.

Incorrect Answers

A:

A prompt-level request cannot override a transaction policy. It does not complete the two decisions required in this question.

C:

Inclusion requires equivalent safety and meaning across supported locales. The stated pair of failures instead requires “Require evidence-backed citation construction and abstention on no results” and “Use layout-preserving extraction and test alternatives”; this option cannot substitute for both.

D:

Threats can enter later stages even when the original user request was harmless. The stated pair of failures instead requires “Require evidence-backed citation construction and abstention on no results” and “Use layout-preserving extraction and test alternatives”; this option cannot substitute for both.

Question 513

A project needs the following AI behavior: A model occasionally produces extra commentary after a machine-readable answer. What should the designer implement?

  1. Specify the output contract and validate returned fields against the schema
  2. Limit retrieval and chat context to the authenticated case scope
  3. Update the configured deployment name and verify compatibility with the SDK
  4. Check regional model availability and data handling requirements before deployment

Explanation

Correct Answer

A:

Model output constraints help downstream parsing but still require validation. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

B:

Source scoping enforces privacy and also improves answer relevance. That is not sufficient here: Model output constraints help downstream parsing but still require validation.

C:

A model family label is not necessarily the concrete deployed endpoint. That is not sufficient here: Model output constraints help downstream parsing but still require validation.

D:

Not every model or feature is available in every location. That is not sufficient here: Model output constraints help downstream parsing but still require validation.

Question 514

An AI operations team is testing two distinct capabilities. Failure (1): The model adds conclusions absent from the source notes. Failure (2): The workflow generates synthetic speech from the audio input. Which TWO specific corrections should the team apply?

  1. Require source-linked summarization evaluation
  2. Use sentiment analysis instead of named-entity extraction
  3. Replace the TTS stage with recognition/transcription
  4. Use layout-aware structured invoice extraction
  5. Use a near-real-time speech translation pipeline

Explanation

Correct Answers

A:

Correct for failure (1). Summarization compresses content without independently validating missing facts. The proposed correction directly addresses that reported defect.

C:

Correct for failure (2). Speech recognition converts speech into words suitable for search. It independently corrects the second defect.

Incorrect Answers

B:

Sentiment analysis infers expressed attitude rather than extracting names. The stated pair of failures instead requires “Require source-linked summarization evaluation” and “Replace the TTS stage with recognition/transcription”; this option cannot substitute for both.

D:

Table extraction associates quantities, descriptions and prices by row. The stated pair of failures instead requires “Require source-linked summarization evaluation” and “Replace the TTS stage with recognition/transcription”; this option cannot substitute for both.

E:

Speech translation spans audio input, semantic transfer and optional output audio. The stated pair of failures instead requires “Require source-linked summarization evaluation” and “Replace the TTS stage with recognition/transcription”; this option cannot substitute for both.

Question 515

A solution team has selected this control: Require a verified source response before presenting financial values. Which observation is the most direct acceptance evidence?

  1. Out-of-scope questions trigger qualified language and reference suggestions
  2. An adversarial page is summarized without executing its prohibited instructions
  3. Eligible memory entries disappear from active stores after the verified request
  4. A simulated timeout yields a failure status without a fabricated numeric balance

Explanation

Correct Answer

D:

A fluent generation cannot substitute for missing authoritative transaction data. The expected acceptance evidence is: A simulated timeout yields a failure status without a fabricated numeric balance. This observation directly tests the stated behavior.

Incorrect Answers

A:

Honesty requires acknowledging when a claim cannot be substantiated. This observation does not establish the required result: A simulated timeout yields a failure status without a fabricated numeric balance.

B:

Lower-trust text is not a privileged command channel. This observation does not establish the required result: A simulated timeout yields a failure status without a fabricated numeric balance.

C:

Accountability requires evidence of actual data lifecycle actions. This observation does not establish the required result: A simulated timeout yields a failure status without a fabricated numeric balance.

Question 516

Troubleshooting report: The client sends a pathname as plain text without audio data. Which correction is MOST directly targeted to this failure?

  1. Configure playback routing and compatible output decoding
  2. Pass supported audio content to an audio-capable model
  3. XML-escape text values before synthesis
  4. Apply a confirmation prompt and constrained recognition

Explanation

Correct Answer

B:

This correction directly addresses the reported failure. A text-only request cannot interpret sound content by file name alone. The targeted change is: Pass supported audio content to an audio-capable model.

Incorrect Answers

A:

Successful synthesis does not guarantee local audio is routed to a speaker. In the stated case, troubleshooting report: The client sends a pathname as plain text without audio data; “Configure playback routing and compatible output decoding” does not target that failure.

C:

SSML is XML and malformed markup breaks the synthesis request. In the stated case, troubleshooting report: The client sends a pathname as plain text without audio data; “XML-escape text values before synthesis” does not target that failure.

D:

Speech recognition uncertainty is higher for confusable numeric strings. In the stated case, troubleshooting report: The client sends a pathname as plain text without audio data; “Apply a confirmation prompt and constrained recognition” does not target that failure.

Question 517

An engineering review documents this requirement: A loan advisory model suggests eligibility based on a protected personal attribute. Select the most appropriate solution.

  1. Enforce tenant identity and record authorization before retrieval and extraction
  2. Set explicit uncertainty thresholds and route high-risk decisions to qualified review
  3. Apply retention and deletion controls subject to recorded legal exceptions
  4. Exclude protected proxies from eligibility logic and test the remaining permitted factors

Explanation

Correct Answer

D:

Legitimate credit criteria cannot be replaced by sensitive identity traits. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

Model instructions cannot grant access beyond the underlying data permission boundary. That is not sufficient here: Legitimate credit criteria cannot be replaced by sensitive identity traits.

B:

Safety-critical recommendations require verified constraints outside open-ended generation. That is not sufficient here: Legitimate credit criteria cannot be replaced by sensitive identity traits.

C:

Accountability requires evidence of actual data lifecycle actions. That is not sufficient here: Legitimate credit criteria cannot be replaced by sensitive identity traits.

Question 518

An acceptance tester needs to validate: Evaluate word error rate on the same held-out audio samples. Which observed result proves the intended behavior?

  1. The acronym is spoken according to the approved pronunciation sample
  2. Early spoken words are present in a captured test transcript
  3. Each action item maps to an utterance and assigned speaker
  4. The selected configuration has lower WER on the same dataset

Explanation

Correct Answer

D:

Comparable labeled audio prevents misleading quality claims. The expected acceptance evidence is: The selected configuration has lower WER on the same dataset. This observation directly tests the stated behavior.

Incorrect Answers

A:

SSML can guide pronunciation beyond plain-text guessing. This observation does not establish the required result: The selected configuration has lower WER on the same dataset.

B:

Real-time audio pipelines need coordinated capture and recognition start. This observation does not establish the required result: The selected configuration has lower WER on the same dataset.

C:

Transcription obtains words; summarization extracts decisions and tasks. This observation does not establish the required result: The selected configuration has lower WER on the same dataset.

Question 519

An Azure-based AI team must satisfy this objective: A product catalog needs to suggest semantically related products from written descriptions. Which design choice is best?

  1. Use image segmentation to map visual damage areas instead of assigning one global defect severity
  2. Use text classification with labeled intent examples
  3. Use text embeddings with nearest-neighbor search and business filters
  4. Use clustering with normalized behavioral features

Explanation

Correct Answer

C:

Semantic vectors capture related concepts while filters enforce eligibility. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

Pixel-level region masks localize building damage more precisely than an overall image class label. That is not sufficient here: Semantic vectors capture related concepts while filters enforce eligibility.

B:

Classification maps text to one of several predefined categories. That is not sufficient here: Semantic vectors capture related concepts while filters enforce eligibility.

D:

Clustering discovers groups based on feature similarity. That is not sufficient here: Semantic vectors capture related concepts while filters enforce eligibility.

Question 520

Troubleshooting report: An unsupported claim passes because the style is persuasive. Which correction is MOST directly targeted to this failure?

  1. Reconcile dependent configuration and rerun targeted regressions
  2. Enforce a typed response schema and fail invalid output
  3. Require explicit identifiers and reject invented IDs
  4. Reject claims with no matching source evidence

Explanation

Correct Answer

D:

This correction directly addresses the reported failure. Prompt instructions must be paired with evidence verification for consequential facts. The targeted change is: Reject claims with no matching source evidence.

Incorrect Answers

A:

Environment drift can change behavior without code edits. In the stated case, troubleshooting report: An unsupported claim passes because the style is persuasive; “Reconcile dependent configuration and rerun targeted regressions” does not target that failure.

B:

Model output constraints help downstream parsing but still require validation. In the stated case, troubleshooting report: An unsupported claim passes because the style is persuasive; “Enforce a typed response schema and fail invalid output” does not target that failure.

C:

Prompts can define required preconditions but tools must still enforce them. In the stated case, troubleshooting report: An unsupported claim passes because the style is persuasive; “Require explicit identifiers and reject invented IDs” does not target that failure.

Microsoft Azure AI Fundamentals (AI-901) Practice Tests