Category: AI-900

AI, AI-900, Artificial Intelligence (AI), Generative AI, Microsoft Certification January 31, 2026

Practice Questions: Identify Features of Generative AI Models (AI-900 Exam Prep)

Practice Questions

Question 1

Which scenario is the best example of a generative AI workload?

A. Predicting tomorrow’s temperature based on historical data
B. Classifying emails as spam or not spam
C. Generating a product description from a short prompt
D. Detecting anomalies in server performance metrics

✅ Correct Answer: C

Explanation:
Generative AI models are designed to create new content, such as text, images, or code. Generating a product description is a content creation task, which is a core feature of generative AI.

Question 2

What is a key characteristic that distinguishes generative AI models from traditional machine learning models?

A. They require labeled training data
B. They produce deterministic outputs
C. They generate new data similar to training data
D. They can only be used for classification tasks

✅ Correct Answer: C

Explanation:
Generative AI models learn patterns from data and generate new outputs that resemble the data they were trained on, rather than only predicting labels or numeric values.

Question 3

What role does a prompt play when working with a generative AI model?

A. It retrains the model with new data
B. It defines how the model should generate a response
C. It validates the accuracy of the model
D. It encrypts the generated output

✅ Correct Answer: B

Explanation:
A prompt provides instructions or context that guide the model’s output. It does not retrain the model or affect its underlying parameters.

Question 4

Why can the same prompt sometimes produce different responses from a generative AI model?

A. The model uses rule-based logic
B. The model is deterministic
C. The model generates probabilistic outputs
D. The training data changes after each request

✅ Correct Answer: C

Explanation:
Generative AI models use probabilistic methods, meaning they select likely next outputs rather than fixed responses, which can result in variation.

Question 5

Which feature enables a generative AI model to produce human-like text responses?

A. Feature engineering
B. Context awareness and large-scale pretraining
C. Manual rule definition
D. Binary classification

✅ Correct Answer: B

Explanation:
Generative AI models are pretrained on massive datasets and use context to generate fluent, coherent, human-like responses.

Question 6

Which statement best describes the training approach used by most generative AI models?

A. They are trained only on small, task-specific datasets
B. They are pretrained on large datasets and adapted for many tasks
C. They require real-time retraining for each request
D. They are trained exclusively using reinforcement learning

✅ Correct Answer: B

Explanation:
Generative AI models are typically large pretrained models that can perform multiple tasks without retraining.

Question 7

Which scenario would most likely require the use of a generative AI model?

A. Predicting customer churn
B. Assigning product categories
C. Writing a summary of a long document
D. Detecting fraudulent transactions

✅ Correct Answer: C

Explanation:
Summarization involves creating new text, which is a hallmark of generative AI workloads.

Question 8

What is a common risk associated with generative AI models that requires responsible AI controls?

A. Overfitting to training data
B. Hallucinations and biased outputs
C. Low model accuracy
D. Inability to scale

✅ Correct Answer: B

Explanation:
Generative AI models can produce confident but incorrect information or biased content, making responsible AI safeguards essential.

Question 9

Which feature allows a generative AI model to continue a conversation in a meaningful way?

A. Feature scaling
B. Context retention
C. Label encoding
D. Data normalization

✅ Correct Answer: B

Explanation:
Context retention enables generative AI models to understand previous inputs and generate coherent multi-turn conversations.

Question 10

Which statement best describes the scope of tasks generative AI models can perform?

A. They are limited to a single predefined task
B. They can perform multiple tasks using the same model
C. They must be retrained for each task
D. They only work with numerical data

✅ Correct Answer: B

Explanation:
Generative AI models are general-purpose, capable of handling a wide variety of tasks such as summarization, translation, content generation, and question answering.

Final Exam Tip 💡

If an AI-900 question mentions:

Creating text, images, or code
Prompts
Conversations
Human-like responses

👉 Think: Generative AI model

Go to the AI-900 Exam Prep Hub main page.

AI, AI-900, Artificial Intelligence (AI), Generative AI, Microsoft Certification January 31, 2026

Identify Features of Generative AI Models (AI-900 Exam Prep)

Introduction

Generative AI models are a class of artificial intelligence systems designed to create new content rather than simply analyze or classify existing data. In the AI-900 exam, Microsoft focuses on conceptual understanding, not implementation details. You are expected to recognize what generative AI models do, how they behave, and what makes them different from traditional machine learning models.

Generative AI underpins many modern Azure AI solutions, including Azure OpenAI Service, and plays a central role in text, image, code, and audio generation workloads.

What Is a Generative AI Model?

A generative AI model learns patterns, structure, and relationships from large datasets and uses that knowledge to generate new, original outputs that resemble the data it was trained on.

Unlike predictive models (which output labels or numeric values), generative models produce:

Text
Images
Code
Audio
Synthetic data

Key Features of Generative AI Models (Exam Focus)

1. Content Generation

Generative AI models can create new content rather than selecting from predefined responses.

Examples:

Writing emails, stories, or summaries
Generating images from text descriptions
Producing computer code
Creating conversational responses

AI-900 cue: If the scenario involves creating something new, it likely involves generative AI.

2. Large Pretrained Models

Generative AI models are typically pretrained on massive datasets containing text, images, or other media.

Key characteristics:

Trained on diverse, large-scale data
Capture language structure, context, and semantics
Can generalize to many tasks without retraining

Examples:

Large language models (LLMs)
Multimodal foundation models

3. Prompt-Based Interaction

Generative AI models are commonly controlled using prompts, which are natural language instructions or inputs.

Prompts can:

Ask questions
Provide instructions
Set constraints or styles
Include examples (few-shot prompting)

Exam tip: Prompts guide how the model responds but do not retrain the model.

4. Probabilistic Output (Non-Deterministic)

Generative AI models produce probabilistic responses, meaning:

The same prompt can produce different outputs
Responses are not fixed or guaranteed
Outputs are generated based on likelihood, not rules

This enables creativity but also requires careful validation.

5. Context Awareness

Generative AI models can use context provided in a conversation or prompt to influence responses.

Examples:

Remembering earlier parts of a conversation
Adjusting tone or topic based on prior input
Generating coherent multi-turn dialogue

This is especially relevant for chat-based AI systems.

6. General-Purpose Capability

Generative AI models are often multi-task by design.

A single model can:

Answer questions
Summarize text
Translate languages
Generate explanations
Write code

This contrasts with traditional ML models, which are typically task-specific.

7. Fine-Tuning and Customization

While generative AI models are pretrained, they can be:

Fine-tuned with domain-specific data
Prompt-engineered for specific use cases
Configured with system instructions

For AI-900, it’s important to know customization is possible, not how to implement it.

8. Human-Like Outputs

Generative AI models are designed to produce outputs that appear:

Natural
Fluent
Contextually relevant
Similar to human-generated content

This is especially true for text and conversational AI.

9. Support for Multimodal Data

Some generative AI models can work across multiple data types, such as:

Text → Image
Image → Text
Text → Code

AI-900 expects recognition of this capability, not technical depth.

10. Need for Responsible AI Controls

Generative AI models require safeguards due to risks such as:

Hallucinations (incorrect but confident outputs)
Bias
Harmful or inappropriate content

Microsoft emphasizes:

Content filtering
Responsible AI principles
Human oversight

Generative AI vs Traditional Machine Learning (High-Yield Comparison)

Aspect	Traditional ML	Generative AI
Primary goal	Predict or classify	Create new content
Output type	Labels or numbers	Text, images, code, audio
Task scope	Narrow, specific	Broad, general-purpose
Interaction style	Structured inputs	Natural language prompts
Creativity	None	High

Azure Context (What AI-900 Expects You to Recognize)

Generative AI workloads on Azure are commonly delivered through:

Azure OpenAI Service
Integrated Azure AI tooling
Secure, enterprise-ready AI deployments

You are not expected to know APIs or pricing — only capabilities and use cases.

Common Exam Triggers to Watch For 👀

If a question mentions:

Writing text
Creating images
Generating code
Conversational responses
Prompt-based interaction

➡️ Think: Generative AI model

Summary

For the AI-900 exam, generative AI models are defined by their ability to:

Generate new content
Respond to prompts
Operate probabilistically
Handle multiple tasks
Produce human-like outputs
Require responsible AI safeguards

Understanding these features, not implementation details, is the key to scoring well in this exam section.

Go to the Practice Exam Questions for this topic.

Go to the AI-900 Exam Prep Hub main page.

AI, AI-900, Artificial Intelligence (AI), Generative AI, Machine Learning (ML), Microsoft Certification January 31, 2026January 31, 2026

Generative AI vs Predictive ML vs Traditional AI (AI-900 Exam Prep)

Here is some additional information to help you solidify your knowledge and understanding of the concepts and prep for the AI-900 exam.

Generative AI vs Predictive ML vs Traditional AI comparison matrix

Aspect	Generative AI	Predictive Machine Learning	Traditional (Rule-Based) AI
Primary Purpose	Generate new content	Predict outcomes or values	Execute predefined rules
Typical Output	Text, images, audio, code, video	Labels, categories, numbers, scores	Yes/No decisions or fixed actions
Creates New Content?	✅ Yes	❌ No	❌ No
Learns From Data?	✅ Yes (large-scale pretraining)	✅ Yes (task-specific training)	❌ No (rules written by humans)
Uses Probabilities?	✅ Yes	✅ Yes	❌ No
Deterministic Output?	❌ No (responses may vary)	⚠️ Usually deterministic	✅ Yes
Handles Unstructured Data	✅ Excellent	⚠️ Limited	❌ Poor
Example Tasks	Chatbots, summarization, image generation, translation	Fraud detection, churn prediction, demand forecasting	Eligibility checks, business rules, workflow automation
Typical Algorithms / Models	Transformers, large language models	Regression, classification, clustering models	If-then rules, decision trees (manual)
Training Data Size	Very large, diverse datasets	Moderate, task-specific datasets	None
Needs Prompts?	✅ Yes	❌ No	❌ No
Adaptable to Many Tasks	✅ High	⚠️ Medium	❌ Low
Common Azure Services	Azure OpenAI Service	Azure Machine Learning	Logic Apps, Power Automate
Example Use Case	Generate a marketing email from a prompt	Predict customer churn probability	Approve a loan if all conditions are met

Quick Mental Model / One-Line Summaries

Think of it this way:

Generative AI → “Create something new”
Predictive ML → “Predict or classify something”
Traditional AI → “Follow the rules exactly”

Or put another way:

Generative AI: Produces new content using large pretrained models
Predictive ML: Uses historical data to predict outcomes
Traditional AI: Uses human-defined rules to make decisions

Common AI-900 Trap to Avoid

❌ “Generative AI is just a type of predictive model”

While generative AI uses prediction internally, its goal is content creation, not classification or numeric prediction.

Go to the AI-900 Exam Prep Hub main page.

AI, AI-900, Artificial Intelligence (AI), Generative AI, Machine Learning (ML), Microsoft Certification January 31, 2026

Workload Scenarios to Correct AI Approach mappings (AI-900 Exam Prep)

Here is some additional information to help you solidify your knowledge and prepare for the AI-900 exam.

1. Core AI Approaches

AI Approach	What It’s Best At
Traditional (Rule-Based) AI	Fixed logic, deterministic decisions
Predictive Machine Learning	Predicting values or classifying outcomes
Generative AI	Creating new content from prompts

Another way to relay the same information:

If it follows rules, it’s traditional AI.
If it predicts, it’s ML.
If it creates, it’s generative AI

2. Scenario-to-Approach Mapping

Business Rules & Automation

Scenario	Correct AI Approach	Why
Approve a loan if income > threshold	Traditional AI	Rule-based, no learning required
Route support tickets based on keywords	Traditional AI	Deterministic logic
Enforce compliance policies	Traditional AI	Rules must be followed exactly

Predictive & Analytical Scenarios

Scenario	Correct AI Approach	Why
Predict customer churn	Predictive ML (Classification)	Binary outcome
Forecast product demand	Predictive ML (Regression)	Numeric prediction
Detect credit card fraud	Predictive ML (Classification)	Probability-based decision
Predict house prices	Predictive ML (Regression)	Continuous value
Segment customers	Predictive ML (Clustering)	Discover groups

Natural Language Processing (NLP)

Scenario	Correct AI Approach	Why
Analyze customer sentiment	Predictive ML (NLP)	Classification of sentiment
Extract key phrases from text	Predictive ML (NLP)	Pattern recognition
Recognize named entities	Predictive ML (NLP)	Identify structured info
Translate text	Generative AI / NLP	Generates new text
Summarize documents	Generative AI	Content creation

Computer Vision

Scenario	Correct AI Approach	Why
Identify objects in an image	Predictive ML (Vision)	Classification/detection
Detect faces in images	Predictive ML (Vision)	Pattern recognition
Read printed text from images (OCR)	Predictive ML (Vision)	Extraction task
Generate images from text	Generative AI	Creates new images

Speech Workloads

Scenario	Correct AI Approach	Why
Convert speech to text	Predictive ML (Speech)	Recognition task
Convert text to speech	Generative AI	Generates audio
Identify spoken language	Predictive ML	Classification

Generative AI Scenarios

Scenario	Correct AI Approach	Why
Generate an email from a prompt	Generative AI	New content
Write code from a description	Generative AI	Content generation
Answer questions conversationally	Generative AI	Dynamic responses
Create images from text prompts	Generative AI	Creative output

3. Azure Service Mapping

Scenario Type	Azure Service
Predictive ML	Azure Machine Learning
NLP (Sentiment, Entities)	Azure AI Language
Speech workloads	Azure AI Speech
Vision workloads	Azure AI Vision
Generative AI	Azure OpenAI Service
Rule-based workflows	Logic Apps / Power Automate

4. Common AI-900 Exam Traps

Trap	Correct Thinking
“Translation is classification”	❌ Translation generates text
“Chatbots are always rule-based”	❌ Modern chatbots use generative AI
“OCR generates text”	❌ OCR extracts existing text
“Generative AI replaces ML”	❌ Different goals

Go to the AI-900 Exam Prep Hub main page.

AI, AI-900, Artificial Intelligence (AI), Generative AI, Microsoft Certification January 31, 2026

Practice Questions: Identify Common Scenarios for Generative AI (AI-900 Exam Prep)

Practice Questions

Question 1

You want to build a solution that can write marketing emails based on a short prompt describing a product.
Which AI approach should you use?

A. Traditional rule-based AI
B. Predictive machine learning
C. Generative AI
D. Computer vision

✅ Correct Answer: C

Explanation:
The solution must create new text content based on prompts. This is a defining characteristic of generative AI, not prediction or rules.

Question 2

A chatbot is required to answer open-ended customer questions using natural, conversational language.
Which scenario does this represent?

A. Sentiment analysis
B. Rule-based automation
C. Generative AI
D. Classification

✅ Correct Answer: C

Explanation:
Conversational assistants that generate dynamic responses are a common generative AI scenario. Rule-based chatbots would rely on predefined responses instead.

Question 3

You need a system that summarizes long legal documents into short executive summaries.
Which type of AI workload is this?

A. Predictive machine learning
B. Generative AI
C. Optical character recognition (OCR)
D. Entity recognition

✅ Correct Answer: B

Explanation:
Summarization involves creating new text that condenses original content, which makes this a generative AI workload.

Question 4

Which task is most appropriate for a generative AI model?

A. Predicting house prices
B. Detecting objects in images
C. Translating text between languages
D. Classifying emails as spam

✅ Correct Answer: C

Explanation:
Translation requires generating new text in another language. The other options are predictive or computer vision tasks.

Question 5

A developer wants to generate Python code from a natural language description of a task.
Which AI capability is being used?

A. Regression
B. Classification
C. Rule-based automation
D. Generative AI

✅ Correct Answer: D

Explanation:
Generating source code from prompts is a classic generative AI use case.

Question 6

Which keyword in an exam question most strongly indicates a generative AI scenario?

A. Predict
B. Classify
C. Detect
D. Generate

✅ Correct Answer: D

Explanation:
Words like generate, create, write, and compose are strong indicators of generative AI scenarios on the AI-900 exam.

Question 7

You need to build a solution that creates images based on a text description.
Which Azure service is most appropriate?

A. Azure AI Vision
B. Azure Machine Learning
C. Azure OpenAI Service
D. Azure AI Language

✅ Correct Answer: C

Explanation:
Azure OpenAI Service is the primary Azure service for generative AI workloads, including text and image generation.

Question 8

Which scenario is NOT a generative AI use case?

A. Writing product descriptions
B. Creating an AI-powered tutor
C. Detecting faces in photos
D. Summarizing meeting notes

✅ Correct Answer: C

Explanation:
Face detection analyzes existing images and does not create new content. It is a computer vision task, not generative AI.

Question 9

A company wants an AI system that rewrites customer emails to sound more polite and professional.
Which AI approach should be used?

A. Predictive machine learning
B. Traditional AI
C. Generative AI
D. Clustering

✅ Correct Answer: C

Explanation:
Rewriting or rephrasing text involves generating new text, which is a generative AI capability.

Question 10

Which statement best describes generative AI?

A. It follows predefined business rules
B. It predicts numeric values
C. It groups similar data points
D. It creates new content based on learned patterns

✅ Correct Answer: D

Explanation:
Generative AI models are designed to create new outputs (text, images, code, audio) based on patterns learned from large datasets.

Quick Exam Tip

If the question asks:

“Will the AI create something new?” → Generative AI
“Will the AI predict or label?” → Predictive ML
“Will the AI follow strict rules?” → Traditional AI

Go to the AI-900 Exam Prep Hub main page.

AI, AI-900, Artificial Intelligence (AI), Generative AI, Microsoft Certification January 31, 2026

Identify Common Scenarios for Generative AI (AI-900 Exam Prep)

Overview

In the AI-900: Microsoft Azure AI Fundamentals exam, generative AI represents a significant and growing focus area. This topic assesses your ability to recognize when generative AI is the appropriate solution and how it differs from traditional AI and predictive machine learning.

Generative AI models are designed to create new content—such as text, images, audio, or code—based on patterns learned from large datasets and guided by user prompts.

This article explains common real-world scenarios where generative AI is used, how those scenarios appear on the AI-900 exam, and how they map to Azure services.

What Makes a Scenario “Generative AI”?

A workload is a generative AI scenario when:

The output is newly generated content, not just a prediction or classification
The model responds to natural language prompts or instructions
The output can vary creatively, even for similar inputs

If the task is to predict, classify, or extract, it is not generative AI. If the task is to create, compose, or generate, it is.

Common Generative AI Scenarios (AI-900 Focus)

1. Text Generation

Scenario examples:

Writing emails, reports, or marketing copy
Drafting blog posts or documentation
Generating summaries from bullet points

Why this is generative AI: The model creates original text based on a prompt rather than selecting from predefined responses.

2. Conversational AI and Chatbots

Scenario examples:

AI-powered customer support chatbots
Virtual assistants that answer open-ended questions
Knowledge assistants that explain concepts conversationally

Why this is generative AI: Responses are dynamically generated and context-aware, rather than rule-based or scripted.

3. Text Summarization

Scenario examples:

Summarizing long documents
Creating executive summaries
Condensing meeting transcripts

Why this is generative AI: The model produces a new, concise version of the original content while preserving meaning.

4. Translation and Language Transformation

Scenario examples:

Translating text between languages
Rewriting text to be simpler or more formal
Paraphrasing content

Why this is generative AI: The output text is newly generated rather than extracted or classified.

5. Code Generation and Assistance

Scenario examples:

Generating code from natural language descriptions
Explaining existing code
Refactoring or optimizing code snippets

Why this is generative AI: The model creates original source code based on intent expressed in a prompt.

6. Image Generation

Scenario examples:

Creating images from text prompts
Generating artwork or design concepts
Producing visual content for marketing

Why this is generative AI: The model synthesizes entirely new images rather than identifying objects in existing ones.

7. Audio and Speech Generation

Scenario examples:

Converting text into natural-sounding speech
Generating voiceovers
Creating spoken responses for virtual assistants

Why this is generative AI: The audio output is generated dynamically from text input.

Azure Services Commonly Used for Generative AI

For the AI-900 exam, generative AI scenarios are most commonly associated with:

Azure OpenAI Service
- Large language models (LLMs)
- Text, code, and image generation
- Conversational AI

Other Azure services (such as Azure AI Speech or Language) may support generative capabilities, but Azure OpenAI Service is the primary service to associate with generative AI workloads.

Generative AI vs Other AI Approaches (Quick Contrast)

Task Type	AI Approach
Predict a value or category	Predictive Machine Learning
Follow predefined rules	Traditional AI
Create new text, images, or code	Generative AI

How This Appears on the AI-900 Exam

On the exam, generative AI scenarios are typically described using words such as:

Generate
Create
Write
Summarize
Compose
Respond conversationally

If the question emphasizes creative or open-ended output, generative AI is likely the correct choice.

Key Takeaways for Exam Day

Generative AI is about creation, not prediction
Outputs are flexible and context-aware
Azure OpenAI Service is the primary Azure service for generative AI
If the output did not previously exist, generative AI is likely the answer

Go to the Practice Exam Questions for this topic.

Go to the AI-900 Exam Prep Hub main page.

AI, AI-900, Artificial Intelligence (AI), Generative AI, Microsoft Certification January 31, 2026

Practice Questions: Identify Responsible AI Considerations for Generative AI (AI-900 Exam Prep)

Practice Exam Questions

Question 1

A company uses a generative AI model to create marketing content. They want to ensure the model does not produce offensive or harmful language.

Which Responsible AI principle is being addressed?

A. Transparency
B. Fairness
C. Reliability and Safety
D. Accountability

Correct Answer: C

Explanation:
Preventing harmful or offensive outputs is a core aspect of reliability and safety, which ensures AI systems behave safely under expected conditions.

Question 2

A chatbot powered by generative AI informs users that responses are created by an AI system and may contain errors.

Which Responsible AI principle does this demonstrate?

A. Privacy and Security
B. Transparency
C. Inclusiveness
D. Fairness

Correct Answer: B

Explanation:
Clearly communicating that content is AI-generated and may be inaccurate supports transparency, helping users understand the system’s limitations.

Question 3

A developer ensures that AI-generated job descriptions do not favor or exclude any gender, ethnicity, or age group.

Which Responsible AI principle is being applied?

A. Accountability
B. Fairness
C. Reliability and Safety
D. Privacy

Correct Answer: B

Explanation:
Avoiding bias and discrimination in generated content aligns with the fairness principle.

Question 4

An organization requires a human reviewer to approve all AI-generated responses before they are published on a public website.

Which Responsible AI principle does this represent?

A. Transparency
B. Reliability and Safety
C. Accountability
D. Inclusiveness

Correct Answer: C

Explanation:
Ensuring humans remain responsible for AI outputs demonstrates accountability.

Question 5

A generative AI system is designed so that user prompts and outputs are not stored or used to retrain the model.

Which Responsible AI principle is primarily addressed?

A. Transparency
B. Privacy and Security
C. Fairness
D. Inclusiveness

Correct Answer: B

Explanation:
Protecting user data and preventing unauthorized use of information supports privacy and security.

Question 6

Which feature in Azure AI services helps prevent generative AI models from producing unsafe or inappropriate content?

A. Model training
B. Content filters
C. Data labeling
D. Feature engineering

Correct Answer: B

Explanation:
Content filters are used to block harmful, unsafe, or inappropriate AI-generated outputs.

Question 7

A generative AI model supports multiple languages and produces accessible text for diverse user groups.

Which Responsible AI principle does this best represent?

A. Fairness
B. Transparency
C. Inclusiveness
D. Accountability

Correct Answer: C

Explanation:
Supporting diverse languages and accessibility aligns with the inclusiveness principle.

Question 8

Which scenario best illustrates a Responsible AI concern specific to generative AI?

A. A model classifies images into categories
B. A model predicts future sales
C. A model generates false but confident answers
D. A model stores structured data in a database

Correct Answer: C

Explanation:
Generative AI can produce hallucinations—incorrect but plausible outputs—which is a key Responsible AI concern.

Question 9

Why is Responsible AI especially important for generative AI workloads?

A. Generative AI requires more computing power
B. Generative AI creates new content that can cause harm if uncontrolled
C. Generative AI only works with unstructured data
D. Generative AI replaces traditional machine learning

Correct Answer: B

Explanation:
Because generative AI creates new content, it can introduce bias, misinformation, or harmful outputs if not properly governed.

Question 10

A company uses Azure OpenAI Service and wants to ensure ethical use of generative AI.

Which action best supports Responsible AI practices?

A. Removing all system prompts
B. Enabling content moderation and human review
C. Increasing model size
D. Disabling user authentication

Correct Answer: B

Explanation:
Combining content moderation with human oversight helps ensure safe, ethical, and responsible use of generative AI.

Final Exam Tips for This Topic

Expect scenario-based questions
Focus on principles, not technical configuration
Watch for keywords: bias, harm, safety, privacy, transparency
If the question mentions risk or trust, think Responsible AI

Go to the AI-900 Exam Prep Hub main page.

AI, AI-900, Artificial Intelligence (AI), Generative AI, Microsoft Certification January 31, 2026

Identify Responsible AI Considerations for Generative AI (AI-900 Exam Prep)

Overview

Generative AI systems are powerful because they can create new content, such as text, images, code, and audio. However, this power also introduces ethical, legal, and societal risks. For this reason, Responsible AI is a core concept tested in the AI-900 exam, especially for generative AI workloads on Azure.

Microsoft emphasizes Responsible AI to ensure that AI systems are:

Fair
Reliable
Safe
Transparent
Secure
Inclusive
Accountable

Understanding these principles — and how they apply specifically to generative AI — is essential for passing the exam.

What Is Responsible AI?

Responsible AI refers to designing, developing, and deploying AI systems in ways that:

Minimize harm
Promote fairness and trust
Respect privacy and security
Provide transparency and accountability

Microsoft has formalized this through its Responsible AI Principles, which are directly reflected in Azure AI services and exam questions.

Why Responsible AI Matters for Generative AI

Generative AI introduces unique risks, including:

Producing biased or harmful content
Generating incorrect or misleading information (hallucinations)
Exposing sensitive or copyrighted data
Being misused for impersonation or misinformation

Because generative AI creates content dynamically, guardrails and safeguards are critical.

Microsoft’s Responsible AI Principles (Exam-Relevant)

1. Fairness

Definition:
AI systems should treat all people fairly and avoid bias.

Generative AI Example:
A text-generation model should not produce discriminatory language based on race, gender, age, or religion.

Azure Support:

Bias evaluation
Content filtering
Prompt design best practices

Exam Clue Words: bias, discrimination, fairness

2. Reliability and Safety

Definition:
AI systems should perform consistently and safely under expected conditions.

Generative AI Example:
A chatbot should avoid generating dangerous instructions or harmful advice.

Azure Support:

Content moderation
Safety filters
System message controls

Exam Clue Words: safety, harmful output, reliability

3. Privacy and Security

Definition:
AI systems must protect user data and respect privacy.

Generative AI Example:
A model should not store or reveal personal or confidential information provided in prompts.

Azure Support:

Data isolation
No training on customer prompts (Azure OpenAI)
Enterprise-grade security

Exam Clue Words: privacy, personal data, security

4. Transparency

Definition:
Users should understand how AI systems are being used and their limitations.

Generative AI Example:
Informing users that responses are AI-generated and may contain errors.

Azure Support:

Model documentation
Clear service descriptions
Usage disclosures

Exam Clue Words: explainability, transparency, disclosure

5. Accountability

Definition:
Humans must remain responsible for AI system outcomes.

Generative AI Example:
A human reviews AI-generated content before publishing it externally.

Azure Support:

Human-in-the-loop design
Monitoring and logging
Responsible deployment guidance

Exam Clue Words: human oversight, accountability

6. Inclusiveness

Definition:
AI systems should empower everyone and avoid excluding groups.

Generative AI Example:
Supporting multiple languages or accessibility-friendly outputs.

Azure Support:

Multilingual models
Accessibility-aware services

Exam Clue Words: inclusivity, accessibility

Responsible AI Controls for Generative AI on Azure

Azure provides built-in mechanisms to help organizations use generative AI responsibly.

Key Controls to Know for AI-900

Control	Purpose
Content filters	Prevent harmful, unsafe, or inappropriate outputs
Prompt engineering	Guide model behavior safely
System messages	Set boundaries for AI behavior
Human review	Validate outputs before use
Usage monitoring	Detect misuse or anomalies

Common Responsible AI Scenarios (Exam Focus)

You are very likely to see scenarios like these:

Preventing a chatbot from generating offensive language
Ensuring AI-generated content is reviewed by humans
Avoiding bias in generated job descriptions
Protecting personal data in prompts and outputs
Informing users that AI-generated content may be inaccurate

If the question mentions risk, harm, bias, safety, or trust, it is almost always testing Responsible AI.

Generative AI vs Responsible AI (Exam Framing)

Concept	Purpose
Generative AI	Creates new content
Responsible AI	Ensures that content is safe, fair, and trustworthy

👉 Generative AI answers what AI can do
👉 Responsible AI answers how AI should be used

Key Takeaways for the AI-900 Exam

Responsible AI is not optional — it is a core design principle
Generative AI introduces new ethical risks
Microsoft’s Responsible AI principles guide Azure AI services
Expect scenario-based questions, not deep technical ones
Focus on concepts, not implementation details

Go to the Practice Exam Questions for this topic.

Go to the AI-900 Exam Prep Hub main page.

AI, AI-900, Artificial Intelligence (AI), Microsoft Certification January 31, 2026

Practice Questions: Describe Features and Capabilities of Azure AI Foundry (AI-900 Exam Prep)

Practice Questions

Question 1

What is the primary purpose of Azure AI Foundry?

A. To provide pre-trained computer vision models only
B. To host virtual machines for AI workloads
C. To provide a unified platform for building, customizing, and managing generative AI solutions
D. To replace Azure Machine Learning

✅ Correct Answer: C

Explanation:
Azure AI Foundry is a unified platform designed to help teams build, customize, deploy, and manage generative AI applications at scale. It does not replace Azure ML but complements it.

Question 2

Which capability of Azure AI Foundry allows organizations to compare and select the most appropriate model for a specific use case?

A. Role-Based Access Control (RBAC)
B. Model catalog and benchmarking
C. Azure Monitor integration
D. Speech synthesis APIs

✅ Correct Answer: B

Explanation:
Azure AI Foundry includes a model catalog with tools to compare and benchmark multiple models, helping teams choose the best model based on performance, cost, or task suitability.

Question 3

A development team wants to create an AI system that can autonomously perform tasks and collaborate with other AI components. Which Azure AI Foundry capability supports this scenario?

A. Image classification
B. Agent orchestration
C. Text analytics
D. Speech recognition

✅ Correct Answer: B

Explanation:
Azure AI Foundry supports AI agents and multi-agent workflows, enabling autonomous task execution and collaboration across agents.

Question 4

Which feature makes Azure AI Foundry suitable for enterprise environments?

A. Open-source licensing
B. Built-in gaming engines
C. Governance, monitoring, and role-based access controls
D. Support for only a single AI model

✅ Correct Answer: C

Explanation:
Enterprise readiness comes from security, governance, RBAC, monitoring, and compliance controls, all of which are core features of Azure AI Foundry.

Question 5

Which task can be performed using Azure AI Foundry?

A. Only training custom neural networks from scratch
B. Managing physical AI hardware
C. Fine-tuning generative AI models for domain-specific use cases
D. Replacing Azure App Service

✅ Correct Answer: C

Explanation:
Azure AI Foundry allows fine-tuning and optimization of generative AI models to adapt them to specific business or domain requirements.

Question 6

What stage of the AI lifecycle is supported by Azure AI Foundry?

A. Only model training
B. Only deployment
C. Only monitoring
D. The full lifecycle from experimentation to production and monitoring

✅ Correct Answer: D

Explanation:
Azure AI Foundry supports the entire AI lifecycle, including experimentation, development, deployment, monitoring, and continuous improvement.

Question 7

Which scenario best matches the use of Azure AI Foundry?

A. Classifying images of animals
B. Translating text between languages
C. Building an enterprise chatbot that uses multiple AI models and enforces governance
D. Running batch SQL queries

✅ Correct Answer: C

Explanation:
Azure AI Foundry is designed for complex generative AI scenarios, such as enterprise chatbots that require multiple models, orchestration, and governance.

Question 8

How does Azure AI Foundry integrate with other Azure services?

A. It operates completely independently
B. It only integrates with Azure OpenAI
C. It integrates with services like Azure App Service, Cosmos DB, and Logic Apps
D. It replaces all other Azure AI services

✅ Correct Answer: C

Explanation:
Azure AI Foundry integrates deeply with the Azure ecosystem, allowing generative AI solutions to be embedded into broader applications and workflows.

Question 9

Which feature helps control access and usage of AI resources in Azure AI Foundry?

A. Prompt engineering
B. Role-Based Access Control (RBAC)
C. Image tagging
D. Speech transcription

✅ Correct Answer: B

Explanation:
RBAC ensures that users and teams only have access to the resources and actions they are authorized to use, supporting secure enterprise deployments.

Question 10

On the AI-900 exam, when should you select Azure AI Foundry as the correct answer?

A. When the question focuses on basic image processing
B. When the question mentions simple sentiment analysis
C. When the scenario describes building, managing, and governing generative AI applications at scale
D. When the question requires only translation services

✅ Correct Answer: C

Explanation:
Azure AI Foundry is the best choice when the scenario involves enterprise-scale generative AI, including model selection, agents, lifecycle management, and governance.

Quick Exam Summary

If the question mentions:

Generative AI
Multiple models
Agents or workflows
Enterprise governance
End-to-end AI lifecycle

👉 Think: Azure AI Foundry

Go to the AI-900 Exam Prep Hub main page.

AI, AI-900, Artificial Intelligence (AI), Microsoft Certification January 31, 2026

Describe Features and Capabilities of Azure AI Foundry (AI-900 Exam Prep)

What Is Azure AI Foundry?

Azure AI Foundry — now commonly referred to as Microsoft Foundry — is a unified Azure platform for developing, managing, and scaling enterprise-grade generative AI applications. It brings together models, tools, governance, and infrastructure into a single, interoperable environment, making it easier for teams to build, deploy, and operate AI apps and agents securely and consistently.

For AI-900 purposes, think of Foundry as a comprehensive hub for generative AI development on Azure — far beyond just model hosting — that enables rapid innovation with governance and enterprise readiness built in.

Core Capabilities of Azure AI Foundry

📌 1. Unified AI Development Platform

Foundry provides a single platform for AI teams and developers to:

Explore and compare a broad catalog of foundational models
Build, test, and customize generative AI solutions
Monitor and refine models over time

This reduces complexity and streamlines workflows compared with managing disparate tools.

🧠 2. Vast Model Catalog & Interoperability

Foundry gives access to thousands of models from multiple sources:

Frontier and open models from Microsoft
Models from OpenAI
Third-party models (e.g., Meta, Mistral)
Partner and community models

Teams can benchmark and compare models for specific tasks before selecting one for production.

⚙️ 3. Customization and Optimization

Foundry provides tools to help you:

Fine-tune models for specific domain needs
Distill or upgrade models to improve quality or reduce cost
Route workloads to the best performing model for a given request

Automated routing helps balance performance vs cost in production AI applications.

🤖 4. Build Agents and Intelligent Workflows

With Foundry, developers can build:

AI agents that perform tasks autonomously
Multi-agent systems where agents collaborate to solve complex problems
RPA-like automation and AI-driven business logic

These agents can be integrated into apps, bots, or workflow systems to respond, act, and collaborate with users.

🔐 5. Enterprise-Ready Governance and Security

Foundry includes enterprise-grade tools to manage:

Role-Based Access Control (RBAC)
Monitoring, logging, and audit trails
Secure access and isolation between teams
Compliance with organizational policies

This makes it suitable for large teams and critical use cases.

🛠 6. Integrated Tools and Templates

Foundry includes:

Pre-built solution templates for common AI patterns (e.g., Q&A bots, document assistants)
SDKs and APIs for Python, C#, and other languages
IDE integrations (e.g., Visual Studio Code extensions)

These accelerate development and reduce the learning curve.

🔄 7. End-to-End Lifecycle Support

Foundry supports the full AI project lifecycle:

Experimentation with models
Development of applications or workflows
Testing and evaluation
Deployment to production
Monitoring and refinement for optimization

This means teams can start with prototypes and scale seamlessly.

🧩 8. Integration with Azure Ecosystem

Foundry is not limited to AI models — it integrates with other Azure services, such as:

Azure App Service
Azure Container Apps
Azure Cosmos DB
Azure Logic Apps
Microsoft 365 and Teams

This allows generative AI features to be embedded into broader enterprise systems.

Scenarios Where Azure AI Foundry Is Used

Foundry supports many generative AI workloads, including:

Conversational agents and bots
Knowledge-powered search and assistants
Context-aware automation
Enterprise RAG (Retrieval-Augmented Generation)
AI-powered workflows and multi-agent orchestration

Its focus on flexibility and scale makes it suitable for both prototyping and enterprise production.

How Foundry Relates to Other Azure Generative AI Services

Capability	Azure AI Foundry	Other Azure Services
Model hosting & comparison	✅	Azure OpenAI / Azure AI services
Multi-model catalog	✅	Individual service catalogs
Fine-tuning & optimization	✅	Azure Machine Learning
Build agents & workflows	✅	Azure AI Language / Bots
Governance & enterprise features	✅	Core Azure security services
Rapid prototyping templates	✅	Individual service templates

Foundry’s value is in bringing these capabilities together into a unified platform.

Exam Tips for AI-900

Foundry is the answer when a question describes building, customizing, and governing enterprise generative AI solutions at scale.
It is not just a model API, but a platform for development, deployment, and lifecycle management of generative AI apps.
If a question mentions agents, workflows, integrated governance, or multi-model support for generative workloads, think Azure AI Foundry / Microsoft Foundry.

Key Takeaways

Azure AI Foundry (Microsoft Foundry) is a unified enterprise AI platform for generative AI development on Azure.
It provides model catalogs, customization, development tools, agents, governance, and integrations.
It supports the full AI application lifecycle — from prototype to production.
It integrates deeply with the Azure ecosystem and supports enterprise-grade governance and security.

Go to the Practice Exam Questions for this topic.

Go to the AI-900 Exam Prep Hub main page.