Microsoft AI-900 Is Retired: Move to AI-901

Microsoft AI-900 retired on June 30, 2026. Azure AI Fundamentals remains an active certification, but the required exam is now Microsoft AI-901. The new exam is more technical than the old AI-900 path: candidates still need core AI concepts, but they also need foundational implementation awareness, familiarity with Azure resources, and basic Python knowledge.

What changed from AI-900 to AI-901

AI-900 was primarily conceptual. It asked candidates to recognize common AI workloads, machine-learning ideas, computer vision, natural-language processing, generative AI, and responsible AI. AI-901 keeps those foundations but places more weight on implementing AI solutions with Microsoft Foundry. That means a current study plan should include both concepts and simple hands-on work.

Responsible AI still anchors the fundamentals

Fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability are not optional theory. Practice applying them to scenarios. If a model performs differently across user groups, which principle is affected? If sensitive prompts are logged carelessly, what risk is introduced? If users cannot understand why a system made a decision, which governance concern appears? The AI-901 workloads and responsible AI guide is useful for this reasoning.

Learn the building blocks of an AI solution

Know the role of models, endpoints, prompts, data, retrieval, embeddings, and evaluation at a beginner level. You do not need advanced mathematics, but you should understand that model selection, grounding data, configuration, and monitoring influence quality. Study the difference between using a general model and building an application that adds instructions, context, tools, and controls around that model.

Add basic Python and API literacy

AI-901 expects foundational technical skills. Practice reading simple Python, sending a request through an SDK or REST API, handling a response, and understanding where credentials and endpoints belong. The goal is not to become a software engineer before the exam; it is to be able to follow how an application calls an AI service and where failures can occur.

Use Microsoft Foundry as the practical context

Create a small project that lets you deploy or access a model, test prompts, observe outputs, and compare results. Then add a second step such as grounding with data or calling a tool. The AI-901 models and deployment guide can help connect those actions to exam terminology without turning the study plan into menu memorization.

Practice one end-to-end AI decision

Use one scenario to connect the whole syllabus. For example, imagine a support team wants an assistant grounded in internal documentation. Identify the workload, the data that should be available, the privacy risks, the model or capability needed, how the application would call it, and what evidence would show the output is useful and safe. This exercise exposes whether you understand only isolated definitions or can combine responsible AI, application flow, data, prompts, and evaluation into one beginner-level design.

Reuse old AI-900 knowledge carefully

Old AI-900 material is still useful for responsible AI, workload categories, machine learning concepts, vision, language, and generative AI. It is not sufficient by itself because the active exam expects more implementation context. Keep the concepts, then add the technical layer that AI-901 introduced.

Current readiness standard

Prepare for AI-901 until you can explain an AI workload, choose an appropriate capability, identify the main responsible-AI risk, and trace how a simple application connects to an Azure AI resource. That combination of concept and implementation is the clearest difference between the retired AI-900 path and the current fundamentals exam.