Certiva
CertificationsPractice TestsFree QuestionsPremium TestsResources
Certiva

Practice. Prepare. Certify. Original practice tests for professional certifications, independent of every exam vendor.

Platform

  • Catalog
  • Free questions
  • Premium tests
  • How it works

Certifications

  • All certifications
  • Microsoft
  • AWS
  • Practice tests

Resources

  • Study resources
  • FAQ
  • My library

Support

  • Help center
  • Contact
  • Log in

Company

  • About
  • Earn
  • Privacy Policy
  • Terms
  • Refund Policy

Social

  • Instagram
  • LinkedIn

© 2026 Certiva. All rights reserved.

PrivacyTermsRefundsContact
Unlock full PDF

Microsoft : Azure AI Fundamentals Practice

Azure AI Fundamentals Practice · Exam last updated: 19 Sept 2026

Total Questions

20

Download Full PDF →

Question: 1

Certiva

A hospital chatbot must refuse to invent drug dosages. Which principle should the team prioritize?

  • A. Maximize fluency even if facts are guessed
  • B. Require human review before any clinical recommendation is shown
  • C. Disable logging to protect speed
  • D. Train only on social media comments

Answer: B

Explanation:

Safety and human oversight beat a fluent but fabricated dosage. Keep a clinician in the loop.

Question: 2

Certiva

A retailer wants product-image captions generated at scale. Which service style fits that workload?

  • A. A telephony voice gateway
  • B. An image analysis / computer-vision captioning API
  • C. A packet capture appliance
  • D. A hardware TPM only

Answer: B

Explanation:

Vision captioning is an image-understanding job, not a speech-to-text or search-index job.

Question: 3

Certiva

A model scores loan applications. What is the most responsible next step before launch?

  • A. Hide the model from auditors
  • B. Evaluate fairness metrics and document human appeal paths
  • C. Use only the highest-risk features
  • D. Skip monitoring after go-live

Answer: B

Explanation:

Measure disparate impact and document the decision policy. Shipping untested scoring is not responsible AI.

Question: 4

Certiva

Users dictate notes on a factory floor. Which capability matches the constraint?

  • A. Offline tape backup only
  • B. Speech-to-text transcription
  • C. A content delivery network
  • D. A barcode printer

Answer: B

Explanation:

Speech-to-text is the matching service when the input is audio and the output is text.

Question: 5

Certiva

A support bot must not reveal another customer’s ticket. Which control is essential?

  • A. Increase temperature for creativity
  • B. Enforce tenant-scoped retrieval before the model answers
  • C. Share one prompt cache across all tenants
  • D. Disable authentication

Answer: B

Explanation:

Authorization and data isolation prevent cross-tenant leakage, even if the model is fluent.

Question: 6

Certiva

A team wants to classify support emails into billing vs. outage. What is the simplest fit?

  • A. A GPU reservation with no labels
  • B. A text classification model with labeled examples
  • C. Changing MX records
  • D. A 3D renderer

Answer: B

Explanation:

Text classification maps labeled categories. It is not a video indexer or a DNS change.

Question: 7

Certiva

Which statement about training data is true for a workplace copilot?

  • A. Copy a competitor’s private corpus
  • B. Use data you are licensed to process and filter secrets before training
  • C. Ignore retention policies
  • D. Store passwords in the prompt

Answer: B

Explanation:

You need rights to the data and a path to remove sensitive records. Scraping random web pages is not a license.

Question: 8

Certiva

An app translates product manuals into five languages. Which capability is the match?

  • A. A BGP peer
  • B. A translation service for text or documents
  • C. A SAN switch
  • D. A raster plotter

Answer: B

Explanation:

Document or text translation is the dedicated capability; a firewall is not.

Question: 9

Certiva

A model might output copyrighted lyrics. What should the product do?

  • A. Always recite the training excerpt
  • B. Apply content filters and refuse copyrighted verbatim output
  • C. Turn filters off for engagement
  • D. Email the full training set to users

Answer: B

Explanation:

Block or cite restricted content and keep a human appeal path. Silent regurgitation is a risk.

Question: 10

Certiva

Which metric best tracks whether a classifier is guessing a rare class?

  • A. Only overall accuracy
  • B. Precision and recall for the rare class
  • C. GPU temperature
  • D. CDN cache hit ratio

Answer: B

Explanation:

Precision/recall on the rare class beats overall accuracy when the set is imbalanced.

Question: 11

Certiva

A kiosk reads handwritten forms. Which Azure-style capability applies?

  • A. SMTP relay
  • B. Optical character recognition on the form image
  • C. Anycast DNS
  • D. A RAID controller

Answer: B

Explanation:

OCR / document intelligence extracts print and handwriting. A load balancer does not.

Question: 12

Certiva

Stakeholders ask why a model denied a claim. What should you provide?

  • A. A screenshot of GPU graphs
  • B. A documented rationale plus a human appeal
  • C. The raw private training row
  • D. No record of the decision

Answer: B

Explanation:

An explanation of features and a human review path. “The neural net said so” is not enough.

Question: 13

Certiva

A chatbot must stay on-policy for a regulated bank. Which pattern helps?

  • A. Unconstrained web browsing for every answer
  • B. Retrieval from an approved knowledge base with refusal on gaps
  • C. Higher temperature by default
  • D. Shared admin tokens in the prompt

Answer: B

Explanation:

Ground answers in approved retrieval sources and refuse when sources are missing.

Question: 14

Certiva

Which data class should never be pasted into an unmanaged prompt?

  • A. Public marketing slogans
  • B. Credentials, government IDs, and protected health details
  • C. The office street address
  • D. A published blog URL

Answer: B

Explanation:

Secrets, government IDs, and health records need a governed path, not a public playground.

Question: 15

Certiva

A team detects drift after a catalog change. What is the first operational step?

  • A. Delete monitoring to save cost
  • B. Compare live metrics to a baseline and re-evaluate the model
  • C. Ship without logs
  • D. Retrain on a single outlier

Answer: B

Explanation:

Monitor, compare against a baseline, and re-evaluate. Ignoring drift lets quality decay.

Question: 16

Certiva

You need to extract key-value pairs from invoices. Which capability fits?

  • A. A syslog forwarder
  • B. Form or document field extraction
  • C. A spanning-tree tweak
  • D. A color profile

Answer: B

Explanation:

Document intelligence / form extraction is built for invoices. A CDN is not.

Question: 17

Certiva

Which human-in-the-loop design is appropriate for refund decisions over a threshold?

  • A. Auto-approve every refund instantly
  • B. Queue high-value refunds for a human approver
  • C. Hide the decision from finance
  • D. Store PAN data in chat logs

Answer: B

Explanation:

High-impact money movement should pause for a reviewer. Full automation of large refunds is risky.

Question: 18

Certiva

A voice agent must detect a wake phrase in noisy audio. What is the matching skill?

  • A. Port mirroring
  • B. Speech recognition with a custom keyword model
  • C. NTP authentication
  • D. A plotter driver

Answer: B

Explanation:

Keyword spotting / speech recognition with noise handling, not a packet sniffer.

Question: 19

Certiva

What is a valid reason to decline generating an image of a real private individual?

  • A. The prompt used complete sentences
  • B. Missing consent to reproduce a real person’s likeness
  • C. The GPU was idle
  • D. The user is on a laptop

Answer: B

Explanation:

Likeness and consent. Public clip-art of objects is a different case.

Question: 20

Certiva

A search experience should rank internal wiki pages. Which pattern applies?

  • A. Open recursive DNS
  • B. Authenticated semantic search over the wiki index
  • C. Public anonymous FTP
  • D. A BIOS update

Answer: B

Explanation:

Semantic or hybrid search over your corpus, with access control, is the fit.

Other exams recommendation

Microsoft : SC-200

Microsoft Security Operations Analyst

View

Microsoft : AZ-900

Azure Fundamentals

View

Microsoft : AZ-104

Azure Administrator

View

Microsoft : MS-900

Microsoft 365 Fundamentals

View

Microsoft : DP-900

Azure Data Fundamentals

View

Microsoft : AI-900

Azure AI Fundamentals

View