Microsoft AI-900 Practice Test

AI-900 practice test: what's covered

• Azure AI Fundamentals

Free AI-900 sample questions

1. A loan-approval model at a bank consistently rejects qualified applicants from one neighborhood at a higher rate than equally qualified applicants elsewhere. Which Responsible AI principle is most directly violated?

   • A. Fairness, because the model treats similar applicants differently based on a protected or proxy attribute
   • B. Transparency, because the model documentation was not published to end users
   • C. Reliability and safety, because the model crashed under unexpected input conditions
   • D. Privacy and security, because applicant data was exposed to unauthorized parties
   Show answer

   Correct: A. Fairness requires that AI treat people in similar situations equitably, so disparate rejection rates for comparable applicants is a fairness violation. Transparency is incorrect because it concerns explainability, not equal treatment. Reliability and safety relates to operating as designed under unexpected conditions, whereas privacy and security concerns data protection, not biased outcomes.

2. A hospital is building an AI triage assistant and wants its design to align with Microsoft's six Responsible AI principles. Which principles map to the listed safeguards? (Choose THREE)

   • A. Inclusiveness, by designing the interface so patients with disabilities can use it
   • B. Accountability, by assigning a governance team responsible for the system's outcomes
   • C. Transparency, by explaining to clinicians how the model reaches its recommendations
   • D. Scalability, by ensuring the system can serve millions of concurrent requests
   • E. Profitability, by maximizing the revenue generated per patient interaction
   Show answer

   Correct: A and B and C. Inclusiveness, accountability, and transparency are three of the six Microsoft Responsible AI principles, alongside fairness, reliability and safety, and privacy and security. Scalability and profitability are engineering and business concerns, not Responsible AI principles, so they are incorrect because Microsoft's framework does not list them.

3. A retailer wants to automatically read printed product labels in photographs, extract the text, and detect the brand logos within each image. Which AI workload category covers this requirement?

   • A. Computer vision, which analyzes images for text, objects, and visual features
   • B. Natural language processing, which interprets and generates human language
   • C. Knowledge mining, which indexes and searches large collections of documents
   • D. Anomaly detection, which flags unusual data points in time-series streams
   Show answer

   Correct: A. Reading text from images (OCR) and detecting logos are computer vision tasks because they extract meaning from visual content. Natural language processing works on language after text exists, not on the image itself. Knowledge mining indexes documents for search, and anomaly detection finds outliers in numeric data, so neither matches image analysis.

4. An autonomous warehouse robot must behave predictably and stop safely when a sensor reports an obstacle it was never trained on. Which Responsible AI principle is the team prioritizing?

   • A. Reliability and safety, ensuring the system responds safely to unanticipated conditions
   • B. Inclusiveness, ensuring the system serves people of all abilities and backgrounds
   • C. Transparency, ensuring stakeholders understand how decisions are made
   • D. Accountability, ensuring people remain responsible for the system's behavior
   Show answer

   Correct: A. Reliability and safety specifically requires systems to operate as designed and respond safely to unexpected situations, which is exactly the obstacle scenario. Inclusiveness concerns serving diverse users and is not about physical safety. Transparency is about explainability and accountability is about human responsibility, so neither directly addresses safe behavior under unforeseen conditions.

5. A data science team wants to predict the exact resale price in dollars of used cars based on mileage, age, and condition. Which type of machine learning problem are they solving?

   • A. Regression, because the target is a continuous numeric value
   • B. Classification, because the target is a discrete category label
   • C. Clustering, because the goal is to group similar records together
   • D. Reinforcement learning, because an agent learns from rewards
   Show answer

   Correct: A. Predicting a continuous numeric outcome such as price is a regression task. Classification is incorrect because it predicts categories like spam or not-spam rather than numbers. Clustering groups unlabeled data and reinforcement learning trains an agent through rewards, so neither produces a specific dollar prediction from labeled examples.

Last updated: 2026-10-19

AI-900 practice test FAQ

Related practice tests

• Azure AZ-900 practice test
• Microsoft DP-900 practice test
• AWS AIF-C01 practice test
• Microsoft SC-900 practice test
• Google Cloud Generative AI Leader practice test
• Microsoft AI-102 practice test