2026 Actual Microsoft AI-901 Dumps, Updated Exam Questions for Guaranteed Pass

AI-901 free sample questions and answers

Microsoft AI-901 practice quiz with correct answers for search indexing.

1. Question 1

   You are building a generative AI customer service chatbot using Azure AI Foundry. You want the model to respond politely but concisely, prioritize resolving common billing issues, and never suggest refunds that exceed company policy. What are the most effective ways to enforce these behaviors? (Select all that apply.)

   AInclude detailed instructions in the system prompt that define the desired tone, priorities, and policy constraints
   

   BRely exclusively on the model's base training to understand company policies and billing procedures
   

   CUse temperature and top-p parameters to control the randomness and diversity of responses
   

   DPost-process model outputs to validate that responses comply with policy before sending to customers
   

   EImplement a separate legal review step that approves every response before it reaches users
   

   Correct answer: A, C, D

   Effective prompt engineering for generative AI involves clear system prompts (to define behavior and constraints), careful tuning of model parameters like temperature and top-p (to control randomness), and validation or filtering of outputs to ensure policy compliance. These are practical, scalable approaches used in production systems.

   Why the other options are incorrect: Relying solely on base training is insufficient because models don't reliably understand specific company policies without explicit guidance; Requiring legal review of every response is impractical and not a standard implementation pattern for customer service chatbots.

2. Question 2

   You have a Microsoft Foundry project that contains an agent named Agent1.

   You need to ensure that Agent1 always calls an Azure function when the agent responds to user input.

   To what should you set tool_choice for Agent1?

   Aauto
   

   Bnone
   

   Crequired
   

   Correct answer: C

   Microsoft's Foundry Agent Service documentation states that tool_choice provides deterministic control over tool calling:

   auto means the model decides whether to call tools. required means the model must call one or more tools. none means the model does not call tools.

   Therefore:

   A . auto = Incorrect, because the model may or may not call the Azure function. B . none = Incorrect, because this prevents tool/function calls. C . required = Correct, because it forces the agent to call a tool.

   The Azure OpenAI function-calling documentation also confirms that tool_choice='auto' lets the model decide whether to call a function, while tool_choice='none' forces a user-facing response without a tool call.

   
   

3. Question 3

   Your organization is developing an AI solution to screen job applications automatically. You want to ensure the model makes fair hiring decisions across all demographic groups. Which of the following actions would best help you address fairness concerns in this AI system?

   ATrain the model on historical hiring data without any modifications to ensure it learns real-world patterns
   

   BAudit the training data for demographic bias, test model predictions across different demographic groups, and establish performance metrics that measure fairness outcomes
   

   CRemove all personally identifiable information from the model to prevent any discrimination
   

   DUse the largest possible dataset available, as larger datasets automatically eliminate bias
   

   EDeploy the model to a limited user group first and monitor only the overall accuracy metric
   

   Correct answer: B

   The correct approach to addressing fairness involves multiple steps: auditing training data for potential bias, testing model performance across demographic groups to identify disparate impacts, and establishing fairness-specific performance metrics—not just overall accuracy. This aligns with responsible AI principles for fairness in AI solutions.

   Why other options are incorrect:
   • Training on unmodified historical data often perpetuates existing societal biases
   • Removing only PII does not address algorithmic bias in decision-making patterns
   • Large datasets do not automatically eliminate bias; they can amplify it if biased patterns exist
   • Monitoring only overall accuracy misses fairness issues that may disproportionately affect specific groups

4. Question 4

   Based on the image provided, here is the transcribed text:

   You need to build an AI solution that produces new product images based on written descriptions provided by users.

   Which AI workload should you use?

   Aimage generation
   

   Bimage analysis
   

   Cobject detection
   

   Doptical character recognition (OCR)
   

   Correct answer: A

   The requirement is to produce new product images based on written descriptions. This is an image generation workload, because the AI system is creating entirely new images from natural language prompts.

   Why the other options are incorrect:

   B . image analysis is used to examine and interpret existing images.

   C . object detection is used to identify and locate objects within an existing image.

   D . optical character recognition (OCR) is used to extract text from images or scanned documents.

   Since the solution must generate new visual content from user-provided descriptions, the correct answer is:

   A . image generation

   
   

5. Question 5

   You have a Microsoft Foundry project that has a generative AI model deployment.

   You need to ensure that responses generated by the model minimize costs and remain within a defined length.

   Which parameter should you configure?

   ATemperature
   

   BMax Completion Tokens
   

   CTop P
   

   DModel version settings
   

   Correct answer: B

   To minimize cost and keep generated responses within a defined length, configure Max Completion Tokens.

   Microsoft's Azure OpenAI / Foundry API reference defines max_completion_tokens as an upper bound for the number of tokens that can be generated for a completion. Because generated tokens contribute to usage and response length, limiting completion tokens helps control both output length and cost.

   Temperature and Top P control randomness or sampling behavior, not maximum response length. Model version settings do not directly define the generated response length.