Microsoft · AI-901
Microsoft Azure AI Fundamentals (AI-901) Practice Test
Validates foundational knowledge of AI concepts and the technical skills to implement AI solutions using Microsoft Azure. Candidates demonstrate understanding of AI workloads, machine learning principles, computer vision, NLP, and generative AI on Azure, with working knowledge of Python and Azure resources.
Exam Details
Questions
600
Duration
60 minutes
Passing Score
700/1000
Difficulty
FoundationalLast Updated
Jun 2026
Topics Covered
Identify AI concepts and responsibilitiesImplement AI solutions using Microsoft FoundryFundamental principles of machine learning on AzureComputer vision workloads on AzureNatural Language Processing (NLP) workloads on AzureGenerative AI workloads on Azure
Microsoft Azure AI Fundamentals (AI-901) Practice Exam Preparation
Use this AI-901 practice exam to prepare for Microsoft Azure AI Fundamentals (AI-901) with realistic questions, detailed explanations, and focused study modes. The practice bank includes 600 questions for Microsoft AI-901, so you can review the exam steadily instead of relying on one long cram session.
As you practice, pay extra attention to recurring topics such as Identify AI concepts and responsibilities, Implement AI solutions using Microsoft Foundry, Fundamental principles of machine learning on Azure, Computer vision workloads on Azure, and Natural Language Processing (NLP) workloads on Azure. Start with short sessions to identify weak areas, then move into timed quizzes once your accuracy is consistent.
The explanations are especially useful when you want to connect exam wording to the responsibilities and scenarios described in the official certification guidance. Use the free preview first, then unlock the full question bank when you are ready to build a complete study routine.
Exam Domain Breakdown
Identify AI Concepts and Responsibilities43%
Implement AI Solutions Using Microsoft Foundry58%
Exam Overview
The Microsoft Azure AI Fundamentals certification (AI-901) validates foundational knowledge of artificial intelligence concepts and the practical technical skills to implement AI solutions using Microsoft Azure. Unlike its predecessor (AI-900), which focused on conceptual understanding, AI-901 reflects the current state of AI development by emphasizing hands-on implementation through Microsoft Azure AI Foundry — Microsoft's unified platform for building, deploying, and managing AI applications. The exam covers a broad spectrum of AI workloads including generative AI, agentic AI, computer vision, natural language processing, speech, and information extraction.
Updated as of April 15, 2026, AI-901 is the replacement for the retiring AI-900 exam (which retires June 30, 2026). Candidates are assessed across two core domains: identifying AI concepts and responsibilities (40–45%) and implementing AI solutions using Microsoft Foundry (55–60%). The implementation-heavy structure means candidates must be comfortable with Python coding syntax, the Azure AI Foundry SDK, and Azure resource management — a notable shift from the purely conceptual nature of the previous exam.
Who Should Take This Exam
AI-901 is designed for individuals at the beginning of their career in AI solution development who want to demonstrate both conceptual knowledge of AI and the foundational technical skills to work with Azure AI services. It suits a wide range of roles including aspiring AI engineers, developers, data analysts, IT professionals, and even business decision-makers with a technical bent who want to understand what Azure AI can do in practice.
The certification also serves as a stepping stone for those pursuing more advanced Microsoft credentials such as Azure AI Engineer Associate or Azure Data Scientist Associate — though it is not a formal prerequisite for those exams. Candidates from non-engineering backgrounds can pursue it, provided they are willing to acquire basic Python familiarity and hands-on Azure experience before the exam.
Prerequisites
There are no mandatory prerequisites to register for or sit the AI-901 exam. However, Microsoft recommends that candidates have awareness of basic cloud concepts and how client-server applications work. Because AI-901 has a substantial implementation domain (55–60% of the exam), candidates should also have working knowledge of Python coding syntax and programming techniques, as well as familiarity with Azure resources and the Azure portal.
Familiarity with REST APIs, SDKs, and CLIs is explicitly called out in the official study guide as expected background knowledge. Candidates who lack hands-on Azure experience are strongly encouraged to complete Microsoft's self-paced learning paths on Microsoft Learn and the official instructor-led course (AI-901T00-A: Introduction to AI in Azure) before attempting the exam.
Exam Format
The AI-901 exam is a proctored, closed-book assessment delivered online or at a Pearson VUE testing center. Based on the standard Microsoft fundamentals exam format and third-party sources, it consists of approximately 40–60 scored questions covering multiple question types including multiple choice, drag-and-drop, case studies, and scenario-based questions. The time allotment is approximately 45–60 minutes of active exam time (the total appointment window including check-in is longer).
The passing score is 700 out of 1000, consistent with all Microsoft certification exams. Scores are reported on a scaled basis and are available immediately upon completion. The exam is available in English and is being progressively localized into additional languages including Japanese, Chinese (Simplified), Korean, German, French, Spanish, Portuguese (Brazil), Russian, and Indonesian. Candidates whose preferred language is not yet available may request an additional 30 minutes. Microsoft strongly recommends registering with a personal Microsoft account (MSA) rather than an organizational account, as records tied to work/school accounts may be lost if you leave that organization.
Skills Measured
- 1.Domain 1 — Identify AI Concepts and Responsibilities (40–45%): Covers the principles of responsible AI (fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability); how generative AI models work; selecting appropriate AI models based on capabilities and deployment configuration parameters; identifying scenarios for common AI workloads including generative AI, agentic AI, text analysis, speech, computer vision, and information extraction; common text analysis techniques such as keyword extraction, entity detection, sentiment analysis, and summarization; features of speech recognition and synthesis; features of computer vision and image-generation models; and techniques to extract information from text, images, audio, and video.
- 2.Domain 2 — Implement AI Solutions Using Microsoft Foundry (55–60%): Covers hands-on implementation across four areas. Generative AI and agents: creating effective system and user prompts, deploying models in the Azure AI Foundry portal, building lightweight chat client apps using the Foundry SDK, and creating and testing single-agent solutions. Text and speech: building applications with text analysis, responding to spoken prompts via deployed multimodal models, and using Azure AI Speech within Foundry Tools. Computer vision and image generation: interpreting visual input with multimodal models, generating images with generative models, and building vision-capable applications. Information extraction: using Azure Content Understanding in Foundry Tools to extract structured data from documents, forms, images, audio, and video.
Study Tips
- Complete the official Microsoft Learn learning paths mapped to AI-901 — navigate to the exam page at learn.microsoft.com/credentials/certifications/exams/ai-901 and follow the 'Two Ways to Prepare' section for the most current self-paced content aligned to the April 2026 objectives.
- Get hands-on with Azure AI Foundry (ai.azure.com) — since 55–60% of the exam tests implementation, spend time actually deploying models, building lightweight chat apps with the Foundry SDK, and testing agent solutions in the Foundry portal rather than only reading documentation.
- Study the official AI-901 Study Guide at learn.microsoft.com/credentials/certifications/resources/study-guides/ai-901 — the bullet-level objectives under each domain tell you exactly what tasks you may be asked to perform or identify, and are the most reliable indicator of exam scope.
- Build hands-on familiarity with Azure Content Understanding in Foundry Tools — this service (covering document, image, audio, and video extraction) is explicitly listed across multiple sub-objectives in the implementation domain and represents a newer capability that many study guides do not yet cover thoroughly.
- Practice writing Python code using the Azure AI Foundry SDK — the study guide specifically notes that candidates should know Python coding syntax and programming techniques and should be familiar with REST APIs, SDKs, and CLIs, indicating code-reading or code-completion questions are likely.
- Use the Microsoft exam sandbox at aka.ms/examdemo to familiarize yourself with the interface and question types before sitting the real exam, reducing time spent on navigation during the actual test.
- Review Microsoft's Responsible AI principles in depth — with 40–45% of the exam dedicated to AI concepts and responsibilities, all six principles (fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability) must be understood in the context of real AI solution design decisions.
Career Benefits
Earning the Microsoft Certified: Azure AI Fundamentals credential (via AI-901) demonstrates to employers a verified baseline of both AI conceptual knowledge and practical Azure implementation ability — a combination increasingly sought as organizations accelerate AI adoption on the Microsoft platform. The certification is relevant for entry-level roles such as AI developer, cloud solutions associate, AI solutions analyst, and technical roles adjacent to data and AI teams. Because AI-901 involves actual implementation tasks using Azure AI Foundry, it carries more practical signal than purely conceptual fundamentals certifications.
As a fundamentals-level credential, AI-901 is positioned as a launchpad rather than a terminal certification. It provides a natural pathway toward higher-value credentials including Azure AI Engineer Associate (AI-102) and Azure Data Scientist Associate (DP-100), which command significantly higher salaries. According to industry surveys, Azure AI Engineer Associates earn average salaries in the range of $120,000–$160,000 USD in the United States, and holding the fundamentals cert demonstrates commitment to that path. The certification is also eligible for ACE college credit, which may provide academic value for candidates pursuing formal education concurrently.
Sample Questions
5 sample questions with answers and explanations. Start a practice session to test yourself across all 600 questions.
Preview — answers shown1. An international news organization called Fabrikam News needs to transcribe live press conferences held in multiple languages in real-time. The transcription must begin immediately as speech starts and continue uninterrupted throughout each event. They are implementing this using Azure AI Speech in Microsoft Foundry. Which speech mode should they configure? (Select one!)
AText-to-Speech synthesis with multilingual neural voices
BSpeaker recognition to identify and verify each conference participant
CContinuous recognition with real-time Speech-to-Text
DBatch transcription for asynchronous processing of uploaded audio files
Explanation
Continuous recognition mode in Azure AI Speech Speech-to-Text processes incoming audio in real-time across 25+ languages, starting immediately as speech begins and continuing without interruption — exactly what is needed for live press conference transcription where results are needed as speech occurs. Batch transcription is asynchronous: it processes pre-recorded audio files after they are fully uploaded to a storage location, making it unsuitable for live events. Speaker recognition identifies or verifies the identity of who is speaking by their voice characteristics but does not transcribe spoken content into text. Text-to-Speech synthesis is the inverse capability, generating spoken audio from written text rather than converting spoken audio to text.
2. The data science team at Datum Corporation needs to train a machine learning model on a dataset containing sensitive private financial records. Their legal team requires a mathematical guarantee that individual-level data cannot be inferred from the model's outputs or query results. They want to use a Microsoft-co-developed open-source toolkit that implements differential privacy techniques. Which toolkit should they use? (Select one!)
ASmartNoise
BFairlearn
CCounterfit
DAzure AI Content Safety
Explanation
SmartNoise is a differential privacy toolkit co-developed by Microsoft that provides mathematical guarantees that individual data records cannot be reliably inferred from model outputs or query results. It works by introducing calibrated statistical noise into computations, ensuring that the presence or absence of any single individual in the training dataset has a negligible, bounded effect on the outputs — satisfying the legal team's requirement for formal privacy guarantees. Fairlearn is an open-source toolkit for assessing and mitigating fairness disparities across demographic subgroups, not for protecting individual privacy during model training. Counterfit is a cybersecurity testing tool that simulates adversarial attacks against AI models to identify vulnerabilities such as model evasion or data extraction. Azure AI Content Safety filters harmful or inappropriate content in model inputs and outputs but provides no differential privacy protections for training data.
3. A developer at Tailspin Toys is building two AI applications on Microsoft Foundry. The first is a creative bedtime story generator that should produce imaginative, varied narratives with each run. The second is a medication reference lookup tool that must provide precise, consistent, and repeatable information every time. Which two temperature configurations should the developer apply to correctly calibrate each application? (Select two!)
Multiple correct answersASet temperature to 1.5 for the story generator to increase output diversity and creative variation between runs
BSet temperature to 0 for the medication reference tool to produce near-deterministic, consistent responses
CSet temperature to 0 for the story generator to ensure children receive predictable and safe content every time
DSet temperature to 2.0 for the medication reference tool to maximize the thoroughness and length of reference answers
ESet frequency penalty to 2 for both applications to prevent repetitive phrasing in generated outputs
Explanation
Temperature controls the randomness of model token selection during output generation. Setting temperature to 1.5 for the story generator samples from a broad probability distribution, producing highly varied and imaginative outputs—exactly what a creative application requires. Setting temperature to 0 for the medication reference tool causes the model to consistently select the highest-probability next token, producing near-deterministic responses essential for accurate, repeatable medical information where consistency is a safety requirement. Setting temperature to 0 for a creative story generator would produce repetitive, uncreative outputs that defeat the application's purpose. A temperature of 2.0 for a medical reference tool dramatically increases hallucination risk by sampling from extremely low-probability tokens, creating a patient safety hazard. Frequency penalty reduces token repetition within a single output but does not control the creative diversity or response consistency across separate requests the way temperature does.
4. A data science team at Contoso Financial has built a machine learning model that assists loan officers in making credit approval decisions. Regulators have requested that the organization demonstrate transparency in how the model reaches its decisions. A project manager proposes publishing the complete model source code, neural network architecture, and all training data publicly to satisfy this requirement. Why does this proposal reflect a misunderstanding of AI transparency? (Select one!)
ATransparency requires providing meaningful, audience-appropriate explanations of how decisions are made, not disclosing proprietary technical internals
BTransparency is only required for high-risk AI systems and does not apply to credit scoring models
CRegulatory transparency requirements are met exclusively through audit trails and incident reporting logs, not model disclosure
DPublishing model source code and training data satisfies transparency requirements but simultaneously creates intellectual property liability
Explanation
AI transparency means making AI systems understandable so that stakeholders can comprehend how decisions are made, calibrated to what is meaningful and appropriate for each audience. For loan officers, this means feature importance scores or SHAP values explaining why a specific applicant was approved or denied. For regulators, this means model cards, datasheets for datasets, and documentation of training methodology and performance metrics across demographic groups. Transparency does not require disclosing proprietary source code, full model architecture, or complete training datasets. The goal is enabling meaningful understanding at the right level of abstraction, not technical openness. Publishing complete training data would also create serious data privacy violations for individuals whose records are in the dataset. Credit decision models are generally considered high-risk AI applications that are subject to transparency expectations — not exempt from them. Audit trails and incident reporting support the accountability principle, which is a distinct responsible AI principle from transparency.
5. Litware Legal Services processes thousands of confidential client case files each month. Before sharing case summaries with external auditors, the legal team must automatically identify all personal information—including phone numbers, email addresses, and passport numbers—and replace those values with placeholder text so the original sensitive data cannot be recovered from shared documents. Which Azure AI Language capability in Microsoft Foundry should the team use? (Select one!)
APII Detection with redaction to identify personal information and replace detected values with masked placeholder text
BKey Phrase Extraction to identify and surface the most important legal concepts within each case file
CNamed Entity Recognition (NER) to categorize mentions of people, locations, dates, and organizations within case files
DSentiment Analysis with opinion mining to evaluate the tone and identify opinion targets across case documents
Explanation
PII Detection is specifically designed to identify personal information such as phone numbers, email addresses, social security numbers, and passport numbers, and critically supports redaction by replacing detected values with placeholder text—making the original values unrecoverable from the processed output, exactly as the external sharing requirement demands. Named Entity Recognition categorizes general entity types such as people, locations, dates, and organizations and can surface some personal references, but it is not specifically optimized for PII identification and does not include built-in redaction functionality that replaces values rather than simply labeling them. Key Phrase Extraction identifies important concepts and topics within documents but has no capability to detect or redact sensitive personal information. Sentiment Analysis measures the emotional tone of text and identifies opinion targets, which is entirely unrelated to personal information identification or document redaction requirements.
$7.99
One-time access to this exam
Full access to all 600 questions
Or $15/mo for all 253 exams
Detailed explanations
Free preview stays available