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 67

Question 660

An AI operations team is testing two distinct capabilities. Failure (1): A raw ampersand breaks the SSML payload. Failure (2): The assistant guesses English when confidence is low. Which TWO specific corrections should the team apply?

  1. XML-escape text values before synthesis
  2. Keep operation outputs separate and map each to its intended field
  3. Use explicit language confirmation or a support fallback
  4. Correct streaming buffer and session lifecycle
  5. Close the capture stream and update permission controls

Explanation

Correct Answers

A:

Correct for failure (1). SSML is XML and malformed markup breaks the synthesis request. The proposed correction directly addresses that reported defect.

C:

Correct for failure (2). Unrecognized languages must not be silently interpreted as a known one. It independently corrects the second defect.

Incorrect Answers

B:

Sentiment polarity and named entity spans answer different questions. The stated pair of failures instead requires “XML-escape text values before synthesis” and “Use explicit language confirmation or a support fallback”; this option cannot substitute for both.

D:

Real-time audio pipelines need coordinated capture and recognition start. The stated pair of failures instead requires “XML-escape text values before synthesis” and “Use explicit language confirmation or a support fallback”; this option cannot substitute for both.

E:

User control of capture must be enforced at the device or application layer. The stated pair of failures instead requires “XML-escape text values before synthesis” and “Use explicit language confirmation or a support fallback”; this option cannot substitute for both.

Question 661

The approved design action is: Assign each request its own authenticated context and correlate responses to request IDs. Which test result would BEST establish that it worked?

  1. Each cited claim resolves to a real approved source chunk
  2. The agent selects the expected tool for a supported test query
  3. Old and new prompts are scored on the same labeled tasks
  4. Parallel tests return each answer to the correct signed-in requester

Explanation

Correct Answer

D:

Concurrent inference must not mix user data or response ownership. The expected acceptance evidence is: Parallel tests return each answer to the correct signed-in requester. This observation directly tests the stated behavior.

Incorrect Answers

A:

Explicit response structure makes evidence easier to verify. This observation does not establish the required result: Parallel tests return each answer to the correct signed-in requester.

B:

A single agent can orchestrate a bounded tool without a multi-agent system. This observation does not establish the required result: Parallel tests return each answer to the correct signed-in requester.

C:

Anecdotal success on different prompts cannot establish an improvement. This observation does not establish the required result: Parallel tests return each answer to the correct signed-in requester.

Question 662

An engineering review documents this requirement: A finance analytics model produces biased risk scores because its training labels encode historical exclusions. Select the most appropriate solution.

  1. Provide tested style instructions and examples in the prompt before considering tuning
  2. Track deployment versions, sampling parameters and test inputs together
  3. Audit training labels and evaluate corrected outcome criteria
  4. Choose a compatible multimodal model and provide supported text/image inputs

Explanation

Correct Answer

C:

Bias can enter through the targets themselves, not just inputs. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

Clear prompting can control tone without model retraining. That is not sufficient here: Bias can enter through the targets themselves, not just inputs.

B:

Reproducibility depends on more than the visible prompt text. That is not sufficient here: Bias can enter through the targets themselves, not just inputs.

D:

Multimodal models can reason over more than one supported input modality. That is not sufficient here: Bias can enter through the targets themselves, not just inputs.

Question 663

Troubleshooting report: Every low-confidence prediction is treated as a verified fact. Which correction is MOST directly targeted to this failure?

  1. Apply confidence and nonmaximum suppression or instance deduplication
  2. Apply a justified threshold with review for uncertain detections
  3. Ground the alt text in the actual picture
  4. Request a clearer image or report uncertainty

Explanation

Correct Answer

B:

This correction directly addresses the reported failure. Confidence scores require calibration and contextual acceptance policy. The targeted change is: Apply a justified threshold with review for uncertain detections.

Incorrect Answers

A:

Object counts require resolving distinct instances, not trusting one scene label. In the stated case, troubleshooting report: Every low-confidence prediction is treated as a verified fact; “Apply confidence and nonmaximum suppression or instance deduplication” does not target that failure.

C:

Accessible captions should communicate image purpose without invented details. In the stated case, troubleshooting report: Every low-confidence prediction is treated as a verified fact; “Ground the alt text in the actual picture” does not target that failure.

D:

Visual reasoning is not proof of an exact value when axes are unclear. In the stated case, troubleshooting report: Every low-confidence prediction is treated as a verified fact; “Request a clearer image or report uncertainty” does not target that failure.

Question 664

A project needs the following AI behavior: A large source document exceeds a deployed model's context limit. What should the designer implement?

  1. Choose a text-classification or sentiment-analysis capability rather than an image generator
  2. Use clustering or another appropriate unsupervised technique
  3. Use retrieval or chunking to supply only relevant passages
  4. Use a conventional calculation function for exact arithmetic and let the model explain it

Explanation

Correct Answer

C:

A finite context window limits how much material can be processed at once. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

Different model tasks require compatible outputs and labeled evaluation data. That is not sufficient here: A finite context window limits how much material can be processed at once.

B:

Unsupervised learning finds structure without supplied target labels. That is not sufficient here: A finite context window limits how much material can be processed at once.

D:

Generative models are not a substitute for auditable deterministic computation. That is not sufficient here: A finite context window limits how much material can be processed at once.

Question 665

Troubleshooting report: The handler maps every empty transcript to user refusal. Which correction is MOST directly targeted to this failure?

  1. Add a no-speech branch and clarification option
  2. Configure playback routing and compatible output decoding
  3. Choose the correct TTS voice from the supported catalog
  4. Adjust recognition vocabulary and test the target terms

Explanation

Correct Answer

A:

This correction directly addresses the reported failure. An empty recognition result is not semantically equivalent to the word no. The targeted change is: Add a no-speech branch and clarification option.

Incorrect Answers

B:

Successful synthesis does not guarantee local audio is routed to a speaker. In the stated case, troubleshooting report: The handler maps every empty transcript to user refusal; “Configure playback routing and compatible output decoding” does not target that failure.

C:

Voice timbre is a synthesis configuration, not speech recognition output. In the stated case, troubleshooting report: The handler maps every empty transcript to user refusal; “Choose the correct TTS voice from the supported catalog” does not target that failure.

D:

Domain-specific words may require biasing or adaptation beyond a generic recognizer. In the stated case, troubleshooting report: The handler maps every empty transcript to user refusal; “Adjust recognition vocabulary and test the target terms” does not target that failure.

Question 666

Troubleshooting report: The QA set contains no examples involving disputed payments. Which correction is MOST directly targeted to this failure?

  1. Add adversarial and exception-focused cases before promotion
  2. Add auditable explanation records and an appeals workflow
  3. Add independent risk review and verifiable test artifacts
  4. Add usable transcript and caption output

Explanation

Correct Answer

A:

This correction directly addresses the reported failure. A high overall score can conceal severe failure on unusual inputs. The targeted change is: Add adversarial and exception-focused cases before promotion.

Incorrect Answers

B:

Accountability includes traceable responsibility and appropriate contestability. In the stated case, troubleshooting report: The QA set contains no examples involving disputed payments; “Add auditable explanation records and an appeals workflow” does not target that failure.

C:

Self-attestation is not independent proof of regulatory controls. In the stated case, troubleshooting report: The QA set contains no examples involving disputed payments; “Add independent risk review and verifiable test artifacts” does not target that failure.

D:

Accessible information must not depend on hearing audio. In the stated case, troubleshooting report: The QA set contains no examples involving disputed payments; “Add usable transcript and caption output” does not target that failure.

Question 667

The approved design action is: Redact personal fields before telemetry storage and control log access. Which test result would BEST establish that it worked?

  1. A simulated timeout yields a failure status without a fabricated numeric balance
  2. The interface clearly marks AI interaction throughout the session
  3. Logs contain operational metadata but not raw customer numbers
  4. A residency map confirms every protected copy remains in allowed locations

Explanation

Correct Answer

C:

Troubleshooting does not justify reproducing sensitive application records. The expected acceptance evidence is: Logs contain operational metadata but not raw customer numbers. This observation directly tests the stated behavior.

Incorrect Answers

A:

A fluent generation cannot substitute for missing authoritative transaction data. This observation does not establish the required result: Logs contain operational metadata but not raw customer numbers.

B:

Transparency prevents users from falsely assuming they are speaking with a human. This observation does not establish the required result: Logs contain operational metadata but not raw customer numbers.

D:

Privacy and compliance require controlling the full data path. This observation does not establish the required result: Logs contain operational metadata but not raw customer numbers.

Question 668

Troubleshooting report: A low max-output-token configuration cuts off the answer. Which correction is MOST directly targeted to this failure?

  1. Validate the tool call and observe an actual execution receipt
  2. Define allowed JSON keys and validate the returned types
  3. Invoke text-read or document OCR capability
  4. Increase the supported limit within deployment quotas or shorten the requested output

Explanation

Correct Answer

D:

This correction directly addresses the reported failure. Generation stops when an output budget or stop condition is reached. The targeted change is: Increase the supported limit within deployment quotas or shorten the requested output.

Incorrect Answers

A:

Agent workflows can reason about tool selection but tools still need authorization. In the stated case, troubleshooting report: A low max-output-token configuration cuts off the answer; “Validate the tool call and observe an actual execution receipt” does not target that failure.

B:

Natural-language fluency is not a guarantee of machine-readable contract validity. In the stated case, troubleshooting report: A low max-output-token configuration cuts off the answer; “Define allowed JSON keys and validate the returned types” does not target that failure.

C:

OCR produces readable text from pixels. In the stated case, troubleshooting report: A low max-output-token configuration cuts off the answer; “Invoke text-read or document OCR capability” does not target that failure.

Question 669

An Azure AI solution uses the following approved control: Include a concise output schema with valid and invalid examples where appropriate. What observable result should pass QA?

  1. A trace shows the task finished and its output is available
  2. The dashboard identifies the specific prompt revision causing increased cost
  3. Test requests consistently produce parseable objects with required keys
  4. Sensitive fields are blocked from unapproved destinations in negative tests

Explanation

Correct Answer

C:

Explicit examples can improve adherence to bounded output conventions. The expected acceptance evidence is: Test requests consistently produce parseable objects with required keys. This observation directly tests the stated behavior.

Incorrect Answers

A:

Submission is not the same as successful completion. This observation does not establish the required result: Test requests consistently produce parseable objects with required keys.

B:

Usage telemetry reveals context-growth and unexpected repetition. This observation does not establish the required result: Test requests consistently produce parseable objects with required keys.

D:

Tool descriptions cannot override data-loss-prevention boundaries. This observation does not establish the required result: Test requests consistently produce parseable objects with required keys.

Question 670

A project needs the following AI behavior: A team wants a system prompt that constrains a support assistant to approved policies. What should the designer implement?

  1. Store the approved instruction as versioned application configuration
  2. Put persistent task boundaries in a supported high-priority system instruction
  3. State the role, goal, provided context, output requirement and constraints clearly
  4. Create one agent in the Foundry portal with scoped tool definitions

Explanation

Correct Answer

B:

High-priority instructions should describe approved behavior and escalation rules. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

Versioned system messages provide a controlled behavioral baseline. That is not sufficient here: High-priority instructions should describe approved behavior and escalation rules.

C:

Specific instructions reduce ambiguity about what constitutes success. That is not sufficient here: High-priority instructions should describe approved behavior and escalation rules.

D:

A single agent can orchestrate a bounded tool without a multi-agent system. That is not sufficient here: High-priority instructions should describe approved behavior and escalation rules.

Microsoft Azure AI Fundamentals (AI-901) Practice Tests