Microsoft AB-100 Practice Questions

The latest changes and updates from the administration for this exam.

verified

Latest Update: Jun 14 2026

All questions are working fine.

All Questions (358)bookmarkBookmarked (0)

All358Show all accessible questions in this exam.Missed358Questions you haven't attempted yet.Incorrect0Questions you answered incorrectly on your last attempt.

Question 101

Your organization deploys a Dynamics 365 Customer Service agent that retrieves external documents as context for generating responses. A security review reveals the agent is vulnerable to prompt manipulation through both user messages and hidden instructions embedded in retrieved documents. Which Azure AI Content Safety capability should the architect implement to address both attack vectors?

Protected material detection, which blocks copyrighted text and code from appearing in model-generated output

Custom categories API, which trains filters to detect and block organization-specific prohibited content patterns

Prompt Shields, which provides a unified API detecting both user prompt attacks and document attacks

Groundedness detection, which validates model outputs against source materials to prevent hallucinated responses

Question 102

Contoso has 15 Azure subscriptions organized under three management groups and needs to ensure all Azure AI services are deployed only within EU regions. What is the MOST effective way to enforce this using Azure Policy?

Assign the built-in Allowed Locations policy at the parent management group with EU regions as allowed parameter values

Create 15 separate custom policy definitions and assign each one individually to every Azure subscription to restrict AI service locations

Deploy Azure Firewall instances in each subscription to inspect and block API requests to Azure AI service endpoints provisioned outside EU Azure regions

Configure Azure RBAC custom role definitions to remove resource creation permissions from users who are located outside EU geographic regions

Question 103

Contoso uses Dynamics 365 Sales with Dataverse. Users ask the Copilot Studio agent about 'closed deals' but the opportunity table uses a status reason value of 'Won.' The agent consistently returns no relevant results for these queries. What should the architect recommend?

Rename the Dataverse status reason value from 'Won' to 'Closed Deals' directly in the opportunity table schema

Deploy a custom Azure Function to intercept and translate user queries before they reach the Dataverse search

Create a separate custom Dataverse table called 'Closed Deals' that duplicates all won opportunity records

Add glossary terms in Copilot Studio that define 'closed deals' as opportunities with a status reason of Won

Question 104

After completing a batch run in Azure Prompt Flow, you realize you need to evaluate the outputs using a different metric than the one originally applied. What should you do?

Clone the entire prompt flow project into a new workspace and modify the evaluation configuration before rerunning

Export the batch run outputs to a local file and create a separate standard flow to calculate the new metric

Start a new evaluation round from the batch run detail page without rerunning the flow

Delete the completed batch run and resubmit it from scratch with the new evaluation method selected

Question 105

Your organization uses Power Automate desktop flows to automate data entry from AI model outputs. You need structured test cases that validate expected results from these flows. Which approach does Microsoft recommend?

Manually observe each desktop flow execution on screen and document the pass or fail results in a shared spreadsheet after every development iteration

Use the Testing module with Assert and Test a desktop flow actions in Power Automate for desktop

Create a separate cloud flow that calls the desktop flow and uses condition actions to check returned values against hardcoded expected outputs in Dataverse

Export desktop flows as JSON definitions and build custom validation scripts in PowerShell to parse the flow output and verify expected action results

Question 106

Your organization is preparing for Microsoft 365 Copilot deployment for enterprise content management and discovers that many SharePoint sites have overly broad sharing permissions. The organization wants to proceed with deployment while mitigating oversharing risks. What should the architect recommend FIRST?

Enable Restricted SharePoint Search to limit Copilot discovery to curated sites while auditing permissions

Delay the Copilot deployment indefinitely until every SharePoint site has completed a comprehensive permission audit

Revoke all existing SharePoint site permissions and require every site owner to reapply for access before deployment

Disable Microsoft Graph for the tenant so Copilot cannot access any organizational content during the permission review

Question 107

An architect needs to provide the operations team with a unified view of Copilot Studio agent health metrics including total conversations, latency, exceptions, and topic analytics in Application Insights. The view must be customizable with KQL and shareable across the team. What should the architect recommend?

Export all raw telemetry data to CSV files on a scheduled basis and build a custom Excel workbook for the operations team to review

Use the Copilot Studio Dashboard in Azure Workbooks under the Monitoring tab in Application Insights

Use the Copilot Studio built-in analytics page exclusively and grant all operations team members Copilot Studio maker access to view it

Create a custom Power Automate cloud flow that sends daily email summaries of agent telemetry data queried from the Application Insights REST API

Question 108

Your organization is configuring the fairness assessment in the Responsible AI dashboard for a customer churn prediction model used with Dynamics 365 Customer Service. How does the dashboard conceptualize fairness when evaluating model behavior?

Individual fairness, which ensures each individual data point always receives an identical prediction score regardless of feature inputs

Architectural fairness, which confirms that every layer of the neural network architecture contributes equally to the model's final prediction output

Sequential fairness, which verifies that model predictions maintain the same output order across all consecutive inference batch runs

Group fairness, which asks which groups defined by sensitive features are at risk of experiencing harm

Question 109

Your organization is evaluating the total cost of ownership for deploying Azure OpenAI Assistants API with Code Interpreter for a data analytics agent that processes daily reports across 15 business units. Each unit runs analytics in its own thread simultaneously. What cost factor is MOST critical to assess?

Code Interpreter incurs per-session charges beyond token fees and simultaneous threads create separate sessions

The total number of Python libraries pre-installed in the Code Interpreter sandbox determines the licensing cost

Code Interpreter sessions are free of charge and only model inference token costs apply to the solution

Code Interpreter charges a flat monthly subscription fee regardless of the number of threads or sessions created

Question 110

A retail company uses Azure AI Language sentiment analysis to process customer feedback from a Copilot Studio agent. Company policy requires that all potentially negative feedback is flagged for review within 24 hours, even if some non-negative entries are also included. How should the architect configure the analysis?

Set a high confidence threshold of 0.95 on the negative sentiment score to ensure only precise matches are flagged

Increase the positive sentiment threshold to 0.99 so that any entries scoring below this value are treated as negative

Use a lower threshold on the negative sentiment score instead of the overall sentiment label

Flag only entries that receive an overall document-level sentiment label of 'mixed' and review those for negativity

Update History

Update History