AI-901: Azure AI Fundamentals Practice Exam

What the AI-901 exam covers

• Identify AI concepts and capabilities324 questions
• Implement AI solutions by using Microsoft Foundry434 questions

Free AI-901 practice test questions

A sample of 10 questions with answers and explanations. A free account includes a fixed set of 20 practice questions for this exam, retryable anytime. Paid access unlocks the full 758-question bank.

1. Question 1Identify AI concepts and capabilities

   In supervised machine learning, what is the role of a label in a training dataset?

   • AThe known output value the model is trained to predictCorrect
   • BThe input variable the model uses to make a prediction
   • CThe numeric weight assigned to each neuron in a network
   • DThe score that measures how accurate a prediction was
   ✓ Correct answer: A

   In supervised learning each training example pairs input features with a label, which is the correct answer for that example. The model adjusts itself to map features to labels, and after training it predicts labels for new, unlabeled data.

   Why the other options are wrong

   • BThat describes a feature, which is an input variable, not the target output.
   • CWeights are internal model parameters learned during training, not labels in the dataset.
   • DA score measuring how accurate a prediction was is an evaluation metric computed after training; the label is the known target value that prediction is compared against.
2. Question 2Identify AI concepts and capabilities

   A generative AI service applies techniques to reduce the chance that it will produce false statements presented as fact, and it grounds answers in a trusted knowledge source. Which responsible AI principle does reducing these fabrications support?

   • AReliability and safetyCorrect
   • BAccountability
   • CInclusiveness
   • DPrivacy and security
   ✓ Correct answer: A

   Reliability and safety require AI to behave dependably and to minimize harm, which for generative models includes reducing inaccurate or fabricated output. Grounding responses in a trusted source helps ensure the system produces reliable answers. Reducing hallucinations is a recognized reliability concern for generative AI.

   Why the other options are wrong

   • BAccountability concerns human responsibility, not the model's factual reliability itself.
   • CInclusiveness concerns serving diverse users, not accuracy of output.
   • DPrivacy and security concern protecting data, not factual reliability.
3. Question 3Implement AI solutions by using Microsoft Foundry

   A prompt includes the instruction 'You are a helpful travel assistant. Only recommend destinations mentioned in the provided list.' Which prompt-engineering elements does this demonstrate?

   • ASetting a persona and constraining the response with a grounding ruleCorrect
   • BIncreasing the model's context window and embedding dimension
   • CFine-tuning the model on travel data
   • DConfiguring the search index sharding strategy
   ✓ Correct answer: A

   The instruction assigns the model a persona ('helpful travel assistant') and adds a constraint that limits answers to a provided list, which is a grounding rule. Both persona setting and constraint specification are core prompt-engineering techniques. Together they shape behavior without changing the model.

   Why the other options are wrong

   • BPrompt text does not alter the context window or embedding dimension.
   • CThis is prompt instruction, not fine-tuning on training data.
   • DIndex sharding is a search configuration, not part of this prompt.
4. Question 4Implement AI solutions by using Microsoft Foundry

   When you create your first project in Microsoft Foundry using the hub-based approach, which resource is created as its parent if one does not already exist?

   • AA hubCorrect
   • BA subnet
   • CA load balancer
   • DA backup vault
   ✓ Correct answer: A

   In the hub-based model, projects must belong to a hub, so creating a project will also create a hub if none exists. The hub then provides shared settings and connections for that project and any future ones. This ensures every project has a governing parent. Hub-based projects remain supported and are reached through the Foundry (classic) portal, and some scenarios such as open-source model deployments still require a hub; for new work the Foundry resource is the default and its projects need no hub.

   Why the other options are wrong

   • BA subnet is a networking construct and is not the parent container for a project.
   • CA load balancer distributes traffic and is not required as a project's parent.
   • DA backup vault stores backups and is unrelated to the project hierarchy.
5. Question 5Implement AI solutions by using Microsoft Foundry

   What file format is commonly required for the training data used in a Microsoft Foundry fine-tuning job for chat models?

   • AJSONL, with one JSON object per lineCorrect
   • BA single XML document
   • CA binary model checkpoint file
   • DA compiled Python module
   ✓ Correct answer: A

   Fine-tuning training and validation data is typically provided in JSON Lines (JSONL) format, where each line is a separate JSON object representing one training example, such as a chat conversation with messages. This structured format lets the service parse each example independently.

   Why the other options are wrong

   • BA single XML document is not the expected input format for fine-tuning data.
   • CA binary checkpoint is a saved model state, not training input data.
   • DA compiled Python module is code, not a training dataset.
6. Question 6Identify AI concepts and capabilities

   Which statement about autonomy levels in agentic solutions is accurate?

   • AAgents can range in autonomy, from needing human approval to acting fully independentlyCorrect
   • BAn agent's autonomy level is set automatically by the size of the language model that backs it
   • CAn agent must run with full autonomy at all times or it is not considered an agent at all
   • DAn agent's autonomy can only ever be raised and never lowered once the agent is deployed
   ✓ Correct answer: A

   Solutions can place a human in the loop for approvals or let the agent act independently, depending on risk and requirements. This makes autonomy a design choice rather than a fixed property. High-stakes actions often keep a human checkpoint while routine tasks may run automatically.

   Why the other options are wrong

   • BAutonomy is a design choice about how much the agent acts on its own, not something determined by the model's parameter count.
   • CAgents can operate at partial autonomy with human checkpoints; full independence is not required for a system to be an agent.
   • DAutonomy levels can be adjusted in either direction through design and guardrails; they are not one-way after deployment.
7. Question 7Identify AI concepts and capabilities

   An organization wants agents built by different teams and even different vendors to communicate using a common, interoperable protocol. Which emerging standard is designed to expose tools and data to agents in a consistent way?

   • AModel Context Protocol (MCP)Correct
   • BSimple Mail Transfer Protocol (SMTP)
   • CHypertext Transfer Protocol (HTTP) only
   • DFile Transfer Protocol (FTP)
   ✓ Correct answer: A

   Model Context Protocol (MCP) is an open standard that lets applications expose tools, data, and context to AI agents in a uniform way, supporting interoperability across different agents and vendors. It reduces custom integration work by giving agents a common method to discover and call capabilities.

   Why the other options are wrong

   • BSMTP is a protocol for sending email and is not designed for agent tool interoperability.
   • CHTTP is a general web transport protocol, not a standard specifically for exposing tools to agents.
   • DFTP transfers files and has nothing to do with agent context or tool discovery.
8. Question 8Identify AI concepts and capabilities

   Which Azure AI Translator capability lets you translate an entire Word or PDF file while preserving its original layout and formatting?

   • ADocument translationCorrect
   • BText translation
   • CTransliteration
   • DDictionary lookup
   ✓ Correct answer: A

   Azure AI Translator offers document translation, which translates whole documents such as Word, PDF, and PowerPoint files while retaining their original structure and formatting. It works asynchronously with files typically stored in blob storage. This differs from text translation, which handles plain strings only.

   Why the other options are wrong

   • BText translation handles plain text strings and does not preserve document layout.
   • CTransliteration converts text from one script to another, not full document translation.
   • DDictionary lookup returns alternative translations for a word or phrase, not whole documents.
9. Question 9Implement AI solutions by using Microsoft Foundry

   An organization wants one curated set of tools - web search, file search and a custom MCP server - shared across many agents, with authentication and versioning managed centrally. Which Foundry feature provides this?

   • AA toolbox, exposed to agents through one managed MCP endpointCorrect
   • BA separate resource group holding one tool each
   • CA connection defined once and referenced by each agent separately
   • DA single agent that every team must reuse unchanged
   ✓ Correct answer: A

   A toolbox groups tools into a reusable unit that Foundry exposes behind a single managed MCP-compatible endpoint, so any agent or runtime can consume them regardless of framework. Because authentication, governance and versioning are centralized on the toolbox, tools are updated in one place instead of rewiring every agent, and a new version can be tested before it is promoted to default.

   Why the other options are wrong

   • BA resource group is a billing and lifecycle container. It does not present tools to an agent.
   • CA connection links one external resource to a project. It does not group tools or present them to agents behind a single endpoint.
   • DSharing one agent does not share tools between different agents, which is the requirement.
10. Question 10Implement AI solutions by using Microsoft Foundry

    A REST client sends this request to Content Understanding: POST {endpoint}/contentunderstanding/analyzers/invoice-analyzer:analyze Content-Type: application/json { "inputs": [ { "url": "https://storage/blob/invoice1.pdf" } ] } What is the request asking the service to do?

    • ARun the invoice-analyzer against the PDF at the URL and return its fieldsCorrect
    • BCreate and register a brand-new analyzer whose name is set to invoice1.pdf
    • CPermanently delete the invoice-analyzer configuration from the resource
    • DTranslate the linked PDF document into several other target languages
    ✓ Correct answer: A

    The path targets an existing analyzer (invoice-analyzer) with the :analyze action, and the body carries an inputs array holding the URL of the file to process. The service applies that analyzer's schema to the document and returns the extracted field values.

    Why the other options are wrong

    • BAnalyzer creation uses a PUT to define the analyzer, not an :analyze call.
    • CDeleting an analyzer would use an HTTP DELETE, not :analyze.
    • DThe call performs extraction, not translation.

Who this AI-901 practice exam is for

This practice set is for anyone preparing for the AI-901: Azure AI Fundamentals exam at the foundational level - from first-time candidates building a foundation to experienced Microsoft practitioners doing a final review before test day. If you learn best by working through realistic questions and reading why each answer is right or wrong, it is built for you.

How to use this AI-901 practice exam

1. Start with the free sample questions above to gauge your current baseline.
2. Read the full explanation on every question, including why each wrong option is wrong.
3. Track your weak domains and focus your study where you are losing the most marks.
4. Once you are scoring consistently well, take a timed, full-length mock exam.
5. Use your readiness score to decide when you are ready to book the real AI-901 exam.

Related Microsoft resources

• AI-901 study guideKey concepts
• Microsoft practice examsAll Microsoft
• Certification pathWhere this fits
• Best AZ-900 Practice Exams (2026)Comparison
• Certification exam guides & tipsBlog
• Plans & pricingFree & paid
• How these questions are written and reviewedMethodology
• Report a problem with a questionCorrections
• AZ-104 practice examRelated
• AZ-140 practice examRelated
• AZ-305 practice examRelated

AI-901 practice exam FAQ