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 16

Question 150

An engineering review documents this requirement: A model describes a chart but invents values absent from the graph. Select the most appropriate solution.

  1. Choose a crop or resize preserving the requested aspect ratio and subject
  2. Ground interpretation in visible labels and qualify uncertain numerical readings
  3. Provide the actual image to a capable multimodal or OCR pipeline
  4. Detect individual package instances and count nonduplicate boxes

Explanation

Correct Answer

B:

Visual reasoning is not proof of an exact value when axes are unclear. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

Smart cropping differs from simply stretching pixels. That is not sufficient here: Visual reasoning is not proof of an exact value when axes are unclear.

C:

Visual evidence is needed to read sign content. That is not sufficient here: Visual reasoning is not proof of an exact value when axes are unclear.

D:

Object counts require resolving distinct instances, not trusting one scene label. That is not sufficient here: Visual reasoning is not proof of an exact value when axes are unclear.

Question 151

Troubleshooting report: The app returns only visual tags such as paper and desk. Which correction is MOST directly targeted to this failure?

  1. Chunk and retrieve relevant source sections before inference
  2. Start with explicit formatting rules and validate style output
  3. Refresh indexed grounding sources and verify their effective date
  4. Invoke text-read or document OCR capability

Explanation

Correct Answer

D:

This correction directly addresses the reported failure. OCR produces readable text from pixels. The targeted change is: Invoke text-read or document OCR capability.

Incorrect Answers

A:

A finite context window limits how much material can be processed at once. In the stated case, troubleshooting report: The app returns only visual tags such as paper and desk; “Chunk and retrieve relevant source sections before inference” does not target that failure.

B:

Clear prompting can control tone without model retraining. In the stated case, troubleshooting report: The app returns only visual tags such as paper and desk; “Start with explicit formatting rules and validate style output” does not target that failure.

C:

External retrieval updates factual context without retraining model weights. In the stated case, troubleshooting report: The app returns only visual tags such as paper and desk; “Refresh indexed grounding sources and verify their effective date” does not target that failure.

Question 152

Troubleshooting report: The assistant guesses English when confidence is low. Which correction is MOST directly targeted to this failure?

  1. Load the deployed speech resource's region and endpoint
  2. Gate side effects on final recognition and explicit confirmation
  3. Add the suitable pronunciation markup or lexicon
  4. Use explicit language confirmation or a support fallback

Explanation

Correct Answer

D:

This correction directly addresses the reported failure. Unrecognized languages must not be silently interpreted as a known one. The targeted change is: Use explicit language confirmation or a support fallback.

Incorrect Answers

A:

Region or endpoint mismatch prevents requests reaching the correct resource. In the stated case, troubleshooting report: The assistant guesses English when confidence is low; “Load the deployed speech resource's region and endpoint” does not target that failure.

B:

Interim hypotheses can change as more audio arrives. In the stated case, troubleshooting report: The assistant guesses English when confidence is low; “Gate side effects on final recognition and explicit confirmation” does not target that failure.

C:

SSML can guide pronunciation beyond plain-text guessing. In the stated case, troubleshooting report: The assistant guesses English when confidence is low; “Add the suitable pronunciation markup or lexicon” does not target that failure.

Question 153

A solution team has selected this control: Use structured outputs or strict application-side schema validation. Which observation is the most direct acceptance evidence?

  1. The summary references events visible on frames and spoken in the transcript
  2. Unseen examples have measured results independent of training
  3. The holdout report identifies missed positives and false alarms explicitly
  4. Malformed or out-of-range fields fail validation before downstream use

Explanation

Correct Answer

D:

Natural-language fluency is not a guarantee of machine-readable contract validity. The expected acceptance evidence is: Malformed or out-of-range fields fail validation before downstream use. This observation directly tests the stated behavior.

Incorrect Answers

A:

Different media channels provide complementary scene and dialogue evidence. This observation does not establish the required result: Malformed or out-of-range fields fail validation before downstream use.

B:

A model can memorize training examples without generalizing. This observation does not establish the required result: Malformed or out-of-range fields fail validation before downstream use.

C:

High accuracy can conceal failure to detect rare positive events. This observation does not establish the required result: Malformed or out-of-range fields fail validation before downstream use.

Question 154

Troubleshooting report: The app shares every unreviewed generated image automatically. Which correction is MOST directly targeted to this failure?

  1. Route to the text-to-image mode with intended inputs only
  2. Map the task to a whole-image classifier
  3. Gate distribution on the image moderation result
  4. Parse outputs by their documented feature-specific schema

Explanation

Correct Answer

C:

This correction directly addresses the reported failure. Generative media needs policy controls in addition to prompt instructions. The targeted change is: Gate distribution on the image moderation result.

Incorrect Answers

A:

Generation from text is distinct from image editing. In the stated case, troubleshooting report: The app shares every unreviewed generated image automatically; “Route to the text-to-image mode with intended inputs only” does not target that failure.

B:

Image-level classification is distinct from object localization. In the stated case, troubleshooting report: The app shares every unreviewed generated image automatically; “Map the task to a whole-image classifier” does not target that failure.

D:

Tags describe concepts while OCR reports character content. In the stated case, troubleshooting report: The app shares every unreviewed generated image automatically; “Parse outputs by their documented feature-specific schema” does not target that failure.

Question 155

Troubleshooting report: The page's hidden instructions are forwarded as privileged role text. Which correction is MOST directly targeted to this failure?

  1. Freeze representative evaluation cases and compare versions
  2. Filter data at the tool boundary using approved policies
  3. Keep publication behind a separate approved deployment step
  4. Separate content channels and gate sensitive actions at the host

Explanation

Correct Answer

D:

This correction directly addresses the reported failure. External content cannot replace privileged system behavior. The targeted change is: Separate content channels and gate sensitive actions at the host.

Incorrect Answers

A:

Anecdotal success on different prompts cannot establish an improvement. In the stated case, troubleshooting report: The page's hidden instructions are forwarded as privileged role text; “Freeze representative evaluation cases and compare versions” does not target that failure.

B:

Tool descriptions cannot override data-loss-prevention boundaries. In the stated case, troubleshooting report: The page's hidden instructions are forwarded as privileged role text; “Filter data at the tool boundary using approved policies” does not target that failure.

C:

A draft or tested agent is not automatically a publicly released service. In the stated case, troubleshooting report: The page's hidden instructions are forwarded as privileged role text; “Keep publication behind a separate approved deployment step” does not target that failure.

Question 156

An acceptance tester needs to validate: Apply authorized retrieval filters to extracted chunks and cited source text. Which observed result proves the intended behavior?

  1. Each action item traces to spoken source intervals
  2. Authorized structured fields remain but raw IDs do not appear in logs
  3. A reviewer confirms the selected clauses and visible signature region from the source, without claiming a verified legal signature
  4. A restricted user cannot search another tenant's contract

Explanation

Correct Answer

D:

Extracting content does not waive document access controls. The expected acceptance evidence is: A restricted user cannot search another tenant's contract. This observation directly tests the stated behavior.

Incorrect Answers

A:

Transcription and summarization are separate results with different validation. This observation does not establish the required result: A restricted user cannot search another tenant's contract.

B:

Inference outputs and telemetry can contain protected identity information. This observation does not establish the required result: A restricted user cannot search another tenant's contract.

C:

Selection-mark extraction is available with GA document analysis, but dedicated signature-location detection is available only in the 2026-06-01-preview API and cannot establish legal validity. This observation does not establish the required result: A restricted user cannot search another tenant's contract.

Question 157

An Azure AI Fundamentals solution has this requirement: A supplier contract has conflicting amounts between a table and an amendment. Which implementation most directly addresses it?

  1. Create a schema for policy type, coverage amount and effective dates
  2. Use a video-capable Content Understanding analyzer with timestamps and supported fields
  3. Extract currency code and monetary amount as separate validated fields
  4. Detect contradictory extracted values and route uncertain cases for review

Explanation

Correct Answer

D:

A model should not silently choose among competing contractual evidence. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

Custom extraction maps semantically relevant phrases to typed fields. That is not sufficient here: A model should not silently choose among competing contractual evidence.

B:

Multimodal extraction draws on visual and spoken evidence over time. That is not sufficient here: A model should not silently choose among competing contractual evidence.

C:

A currency symbol or amount alone may not establish financial meaning. That is not sufficient here: A model should not silently choose among competing contractual evidence.

Question 158

The approved design action is: Use PII-aware entity recognition with appropriate redaction. Which test result would BEST establish that it worked?

  1. Suggested products are relevant and meet inventory/authorization rules
  2. Names in a labeled sample are detected and redacted
  3. A clause query returns the relevant authorized contract sections
  4. The result contains product boxes in their correct shelf positions

Explanation

Correct Answer

B:

Entity extraction locates sensitive spans for privacy handling. The expected acceptance evidence is: Names in a labeled sample are detected and redacted. This observation directly tests the stated behavior.

Incorrect Answers

A:

Semantic vectors capture related concepts while filters enforce eligibility. This observation does not establish the required result: Names in a labeled sample are detected and redacted.

C:

Semantic search can retrieve conceptually related passages. This observation does not establish the required result: Names in a labeled sample are detected and redacted.

D:

Object detectors return classes and spatial boundaries. This observation does not establish the required result: Names in a labeled sample are detected and redacted.

Question 159

A project needs the following AI behavior: A team must extract totals from multiple invoices in one PDF. What should the designer implement?

  1. Preserve page identifiers and validate page sequence during extraction
  2. Classify or segment the combined PDF into logical documents before extraction
  3. Use an appropriate Content Understanding or invoice-specific analyzer with a typed field schema
  4. Minimize extracted PII and restrict diagnostic payload storage

Explanation

Correct Answer

B:

Document boundaries prevent values from unrelated invoices being joined. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

Page order and references are part of document provenance. That is not sufficient here: Document boundaries prevent values from unrelated invoices being joined.

C:

Document extraction turns unstructured pages into validated structured fields. That is not sufficient here: Document boundaries prevent values from unrelated invoices being joined.

D:

Inference outputs and telemetry can contain protected identity information. That is not sufficient here: Document boundaries prevent values from unrelated invoices being joined.

Question 160

A project needs the following AI behavior: A Python client must call a deployed Foundry model rather than a model catalog entry. What should the designer implement?

  1. Configure the client with the actual resource endpoint and deployment identifier
  2. Check supported context and output limits for the chosen deployment
  3. Update the configured deployment name and verify compatibility with the SDK
  4. Treat user content as untrusted and maintain privileged instruction boundaries

Explanation

Correct Answer

A:

An inferencing request is routed to a concrete deployment configured in the resource. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

B:

Capabilities and limits vary by model deployment. That is not sufficient here: An inferencing request is routed to a concrete deployment configured in the resource.

C:

A model family label is not necessarily the concrete deployed endpoint. That is not sufficient here: An inferencing request is routed to a concrete deployment configured in the resource.

D:

A user cannot use text to elevate permissions beyond the application policy. That is not sufficient here: An inferencing request is routed to a concrete deployment configured in the resource.

Microsoft Azure AI Fundamentals (AI-901) Practice Tests