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 9

Question 80

Troubleshooting report: The agent treats the user text as a new system instruction. Which correction is MOST directly targeted to this failure?

  1. Preserve role separation and tool-side permission checks
  2. Add idempotency keys and query the original transaction state
  3. Version prompt and deployment settings and retain the actual evaluation output
  4. Rotate the key and load credentials through a governed secret mechanism

Explanation

Correct Answer

A:

This correction directly addresses the reported failure. A user cannot use text to elevate permissions beyond the application policy. The targeted change is: Preserve role separation and tool-side permission checks.

Incorrect Answers

B:

A timeout does not prove that the original mutation failed. In the stated case, troubleshooting report: The agent treats the user text as a new system instruction; “Add idempotency keys and query the original transaction state” does not target that failure.

C:

Recorded settings support investigation, but generative models can still vary on an apparently identical rerun. In the stated case, troubleshooting report: The agent treats the user text as a new system instruction; “Version prompt and deployment settings and retain the actual evaluation output” does not target that failure.

D:

Credentials should not appear in repositories, prompt logs or client binaries. In the stated case, troubleshooting report: The agent treats the user text as a new system instruction; “Rotate the key and load credentials through a governed secret mechanism” does not target that failure.

Question 81

An Azure AI Fundamentals solution has this requirement: A company processes scanned documents whose pages may arrive out of order. Which implementation most directly addresses it?

  1. Preserve text spans and associated source regions during OCR
  2. Preserve page identifiers and validate page sequence during extraction
  3. Define a custom analyzer for consistent target field meanings across templates
  4. Use supported speaker/timing evidence and keep uncertain attribution explicit

Explanation

Correct Answer

B:

Page order and references are part of document provenance. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

Source-location evidence helps verify recognition on mixed-layout documents. That is not sufficient here: Page order and references are part of document provenance.

C:

A schema-driven analyzer identifies business fields independent of fixed coordinates. That is not sufficient here: Page order and references are part of document provenance.

D:

Conversation extraction needs more than a single flattened text stream. That is not sufficient here: Page order and references are part of document provenance.

Question 82

A project needs the following AI behavior: A photo-processing app needs a list of visual attributes without sentence-level narration. What should the designer implement?

  1. Check the exact API version and available visual feature vocabulary
  2. Request image tagging for the selected visual features
  3. Provide the actual image to a capable multimodal or OCR pipeline
  4. Use optical character recognition for image text

Explanation

Correct Answer

B:

Tags summarize recognized concepts as labels. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

Feature names and availability vary across versions. That is not sufficient here: Tags summarize recognized concepts as labels.

C:

Visual evidence is needed to read sign content. That is not sufficient here: Tags summarize recognized concepts as labels.

D:

OCR extracts characters visible in pixels. That is not sufficient here: Tags summarize recognized concepts as labels.

Question 83

The approved design action is: Use object detection and count accepted object instances. Which test result would BEST establish that it worked?

  1. The item count reconciles with detected boxes at evaluated confidence
  2. The itinerary can be prepared without a charge until confirmation
  3. Workers without matched helmet detections are flagged for review
  4. Totals, dates and merchant names match receipt evidence

Explanation

Correct Answer

A:

Instance detections support counts where one image contains several objects. The expected acceptance evidence is: The item count reconciles with detected boxes at evaluated confidence. This observation directly tests the stated behavior.

Incorrect Answers

B:

An agent may propose actions but payment must remain permission-controlled. This observation does not establish the required result: The item count reconciles with detected boxes at evaluated confidence.

C:

Detection relates objects to positions for compliance checks. This observation does not establish the required result: The item count reconciles with detected boxes at evaluated confidence.

D:

Images of receipts need OCR/layout extraction before typed business values. This observation does not establish the required result: The item count reconciles with detected boxes at evaluated confidence.

Question 84

Troubleshooting report: The system concatenates all text into one unordered string. Which correction is MOST directly targeted to this failure?

  1. Enable supported layout and field extraction features
  2. Use layout-aware field definitions rather than fixed crop positions
  3. Verify the actual source signature region or use explicitly approved preview detection, never infer signing from a typed name
  4. Enable appropriate diarization and handle ambiguous overlap

Explanation

Correct Answer

A:

This correction directly addresses the reported failure. Ordinary plain-text OCR may lose table relationships and checkbox states. The targeted change is: Enable supported layout and field extraction features.

Incorrect Answers

B:

A schema-driven analyzer identifies business fields independent of fixed coordinates. In the stated case, troubleshooting report: The system concatenates all text into one unordered string; “Use layout-aware field definitions rather than fixed crop positions” does not target that failure.

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. In the stated case, troubleshooting report: The system concatenates all text into one unordered string; “Verify the actual source signature region or use explicitly approved preview detection, never infer signing from a typed name” does not target that failure.

D:

Conversation extraction needs more than a single flattened text stream. In the stated case, troubleshooting report: The system concatenates all text into one unordered string; “Enable appropriate diarization and handle ambiguous overlap” does not target that failure.

Question 85

A project needs the following AI behavior: A finance team needs to process invoices with line-item tables. What should the designer implement?

  1. Use image segmentation to map visual damage areas instead of assigning one global defect severity
  2. Use text-to-speech with a supported voice and accessible controls
  3. Use visual-question answering with a multimodal model
  4. Use invoice-oriented structured extraction with table row preservation

Explanation

Correct Answer

D:

Table extraction associates quantities, descriptions and prices by row. 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: Table extraction associates quantities, descriptions and prices by row.

B:

Speech synthesis converts supplied text into audible output. That is not sufficient here: Table extraction associates quantities, descriptions and prices by row.

C:

Image-and-text reasoning can connect a question with visible scene content. That is not sufficient here: Table extraction associates quantities, descriptions and prices by row.

Question 86

The approved design action is: Apply bounded retries and respect the service's throughput limit. Which test result would BEST establish that it worked?

  1. The thumbnail preserves subject proportions and important content
  2. A supported feature call returns the expected output schema
  3. The test workload recovers without unbounded retry bursts
  4. A labeled test shows the chosen threshold achieves target precision/recall

Explanation

Correct Answer

C:

Too many concurrent requests can exceed resource quota. The expected acceptance evidence is: The test workload recovers without unbounded retry bursts. This observation directly tests the stated behavior.

Incorrect Answers

A:

Smart cropping differs from simply stretching pixels. This observation does not establish the required result: The test workload recovers without unbounded retry bursts.

B:

Feature names and availability vary across versions. This observation does not establish the required result: The test workload recovers without unbounded retry bursts.

D:

Confidence scores require calibration and contextual acceptance policy. This observation does not establish the required result: The test workload recovers without unbounded retry bursts.

Question 87

Troubleshooting report: The team assumes English accuracy proves all language support. Which correction is MOST directly targeted to this failure?

  1. Filter data at the tool boundary using approved policies
  2. Run locale-specific evaluations and fallback for unsupported cases
  3. Attach idempotency keys and verify tool receipts after reconnecting
  4. Benchmark candidate deployments and choose by requirements

Explanation

Correct Answer

B:

This correction directly addresses the reported failure. Language coverage can vary by model and deployment. The targeted change is: Run locale-specific evaluations and fallback for unsupported cases.

Incorrect Answers

A:

Tool descriptions cannot override data-loss-prevention boundaries. In the stated case, troubleshooting report: The team assumes English accuracy proves all language support; “Filter data at the tool boundary using approved policies” does not target that failure.

C:

Client retries should distinguish read requests from side-effecting tool calls. In the stated case, troubleshooting report: The team assumes English accuracy proves all language support; “Attach idempotency keys and verify tool receipts after reconnecting” does not target that failure.

D:

Lower-cost deployment is acceptable only if accuracy and safety floors remain met. In the stated case, troubleshooting report: The team assumes English accuracy proves all language support; “Benchmark candidate deployments and choose by requirements” does not target that failure.

Question 88

A solution team has selected this control: Require verifiable grounding and calibrated uncertainty. Which observation is the most direct acceptance evidence?

  1. The model/deployment capability check and a spoken-audio test confirm that this endpoint supports transcription
  2. Unsupported factual claims are flagged and answered with uncertainty
  3. A regression test records model version and key generation settings
  4. A held-out specialty test improves over prompting and retrieval baselines

Explanation

Correct Answer

B:

Language-model likelihood does not independently establish factual truth. The expected acceptance evidence is: Unsupported factual claims are flagged and answered with uncertainty. This observation directly tests the stated behavior.

Incorrect Answers

A:

Model capabilities must include the required audio modality before the client can obtain speech-to-text results. This observation does not establish the required result: Unsupported factual claims are flagged and answered with uncertainty.

C:

Reproducibility depends on more than the visible prompt text. This observation does not establish the required result: Unsupported factual claims are flagged and answered with uncertainty.

D:

Model specialization may help task behavior but requires measured benefit. This observation does not establish the required result: Unsupported factual claims are flagged and answered with uncertainty.

Question 89

An acceptance tester needs to validate: Use supported SSML phoneme or say-as controls for pronunciation. Which observed result proves the intended behavior?

  1. The transcript covers multiple utterances with correct ordering
  2. The acronym is spoken according to the approved pronunciation sample
  3. Subtitle cues match the spoken phrases at measured times
  4. The translated request preserves numbers and intent

Explanation

Correct Answer

B:

SSML can guide pronunciation beyond plain-text guessing. The expected acceptance evidence is: The acronym is spoken according to the approved pronunciation sample. This observation directly tests the stated behavior.

Incorrect Answers

A:

Long interactions need streaming or segmented recognition over time. This observation does not establish the required result: The acronym is spoken according to the approved pronunciation sample.

C:

Timecodes align text segments to audio playback. This observation does not establish the required result: The acronym is spoken according to the approved pronunciation sample.

D:

Translation requires semantic equivalence and language direction. This observation does not establish the required result: The acronym is spoken according to the approved pronunciation sample.

Question 90

An Azure-based AI team must satisfy this objective: A virtual assistant is deployed to staff without informing them it is AI. Which design choice is best?

  1. Test accuracy on representative lighting and camera conditions and provide a safe fallback
  2. Redact personal fields before telemetry storage and control log access
  3. Disclose the automated nature and supported escalation route in the interface
  4. Treat retrieved content as untrusted evidence and prevent it from changing tool authority

Explanation

Correct Answer

C:

Transparency helps users assess the scope and limitations of the assistance. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

Reliability demands testing the actual operating environment, not just clean samples. That is not sufficient here: Transparency helps users assess the scope and limitations of the assistance.

B:

Troubleshooting does not justify reproducing sensitive application records. That is not sufficient here: Transparency helps users assess the scope and limitations of the assistance.

D:

Lower-trust text is not a privileged command channel. That is not sufficient here: Transparency helps users assess the scope and limitations of the assistance.

Microsoft Azure AI Fundamentals (AI-901) Practice Tests