Amazon Web Services · AIF-C01

AIF-C01 Exam: Complete AWS AI Practitioner Guide

Current AIF-C01 exam guide covering the $100 USD price, 90-minute format, five AI domains, 700 passing score, preparation, and recertification.

Last verified September 6, 2026

AIF-C01 exam at a glance

Facts last verified September 6, 2026

Domain breakdown

Who should take the AIF-C01 exam?

AIF-C01 is for people who use, evaluate, purchase, govern, or discuss AI solutions on AWS but do not necessarily build models. AWS lists business analysts, support staff, marketers, product and project managers, line-of-business leaders, IT managers, and sales professionals as examples.

The target candidate has up to six months of exposure to AI or machine learning on AWS. You should recognize core AWS services, the shared responsibility model, IAM, and pricing concepts. The exam focuses on practical business applications of AI rather than advanced statistics or programming.

Model development, feature engineering, hyperparameter tuning, infrastructure implementation, and mathematical analysis are outside the target role. Those exclusions do not make the exam trivial: you still need to choose suitable AI approaches, understand foundation-model applications, assess risk, and identify secure AWS capabilities.

Skills measured on AIF-C01

The official AIF-C01 exam guide defines five domains. Applications of foundation models is the largest at 28%, while responsible AI plus security and governance account for another 28%. A preparation plan focused only on prompt engineering misses more than half the blueprint.

Fundamentals of AI and ML (20%)

Know the relationships among artificial intelligence, machine learning, deep learning, neural networks, natural language processing, computer vision, and generative AI. Distinguish supervised, unsupervised, and reinforcement learning and recognize common tasks such as classification, regression, clustering, forecasting, recommendation, and anomaly detection.

Connect use cases to outcomes. Predictive ML estimates or classifies from patterns, while GenAI creates new content. Understand when rules or conventional analytics are more appropriate than AI. Consider data availability, quality, cost, latency, explainability, and business value before choosing a technique.

Learn lifecycle concepts from data collection and preparation through training, evaluation, deployment, inference, monitoring, and retraining. Recognize accuracy, precision, recall, F1, business metrics, drift, and overfitting at a conceptual level. Associate managed capabilities with services such as Amazon SageMaker AI, Transcribe, Translate, Comprehend, Lex, and Polly.

Fundamentals of generative AI (24%)

Foundation models are large reusable models adaptable to many tasks. Understand tokens, context windows, embeddings, vector representations, inference, multimodal inputs, transformers, diffusion, and how large language models generate output. You do not need to derive the mathematics.

Benefits include adaptable natural-language interfaces, content generation, summarization, classification, extraction, and faster prototyping. Limitations include hallucination, nondeterminism, bias, data and intellectual-property risk, limited context, latency, and cost. A convincing answer may still be factually wrong, so verification and grounding matter.

Compare approaches at a high level: use a model as-is, improve instructions with prompting, ground it with retrieval-augmented generation, customize through fine-tuning, or train a model only when requirements justify the data, cost, and expertise. Know that model choice depends on modality, quality, context, latency, security, licensing, and price.

Applications of foundation models (28%)

This largest domain turns concepts into solution choices. Prompt engineering includes clear instructions, context, examples, constraints, roles, negative instructions, and output formats. Recognize zero-shot and few-shot prompting, prompt templates, and the need to test prompts against representative inputs rather than one successful example.

Retrieval-augmented generation finds relevant external content and supplies it to a model at inference time. Embeddings and vector stores enable semantic retrieval. RAG can improve freshness and grounding without changing model weights, but poor source data, chunking, retrieval, or access control still produces poor results.

Understand Amazon Bedrock as a managed service for building generative AI applications with foundation models. Study model evaluation, Knowledge Bases, Agents, Guardrails, prompt management, and related security features at a capability level. Compare Bedrock with SageMaker AI based on how much model-building and infrastructure control a team requires.

Evaluation should include technical and business criteria: factuality, relevance, robustness, toxicity, bias, latency, cost, and user outcomes. Human evaluation remains important for subjective quality and high-impact decisions.

Guidelines for responsible AI (14%)

Responsible AI considers fairness, safety, privacy, inclusivity, transparency, explainability, accountability, robustness, and human oversight throughout the lifecycle. A policy document alone is insufficient; teams need representative data, risk assessment, testing, monitoring, feedback, incident handling, and defined owners.

Bias can enter through data selection, labels, measurement, objectives, deployment context, or feedback loops. Mitigation may involve better datasets, testing across groups, documented limits, guardrails, and human review. Explainability helps stakeholders understand relevant factors, but the suitable level depends on the use case and model.

Security, compliance, and governance for AI solutions (14%)

Apply least privilege, encryption, network isolation, logging, data classification, and the AWS shared responsibility model. Understand why prompts, retrieved documents, outputs, embeddings, and logs may all contain sensitive data. IAM roles and policies control access; KMS supports key management; PrivateLink can provide private connectivity for supported services.

Governance includes approved use cases and models, inventories, documentation, evaluation gates, retention, monitoring, auditability, and incident response. Regulations vary by jurisdiction and industry, so the exam tests recognition of compliance considerations rather than legal advice.

How to prepare for AIF-C01

Begin with AWS's exam-prep plan and blueprint. Create a concept map that connects each business problem to a technique, service, risk, metric, and control. This prevents shallow memorization—for example, a RAG application also needs authorized retrieval, grounded evaluation, monitoring, and cost management.

Use Amazon Bedrock in a controlled lab if available. Compare models, test prompt variants, inspect token use, create a small knowledge base, and apply guardrails. Record where an output fails and which mitigation addresses the failure. Keep all data synthetic and set a budget.

Practice reading scenario constraints. Words such as current proprietary documents suggest RAG; consistent organization-specific behavior may suggest customization; minimal operational effort suggests managed services; regulated data raises isolation, access, retention, and audit concerns.

AIF-C01 practice questions

High-quality questions ask for the best technique or control under stated constraints. Explain why each distractor fails on accuracy, risk, cost, latency, data requirements, or operational effort. Include ordering and matching practice because AWS lists those item types for AIF-C01.

Use only objective-aligned material. Dumps are unauthorized, may be wrong after updates, and do not teach the judgment needed to evaluate an AI use case.

AIF-C01 compared with Cloud Practitioner

AIF-C01 focuses on AI, ML, GenAI, responsible use, and AI security. CLF-C02 covers AWS Cloud value, core services, general security, operations, billing, and support. Neither requires the other; choose based on whether you need AI literacy or broad cloud literacy. Earning both can demonstrate the intersection without replacing hands-on experience.

Career value of AWS AI Practitioner

The credential is relevant to cloud-adjacent roles that influence AI projects. It signals that you can discuss models, use cases, evaluation, responsible AI, and AWS controls with business and technical stakeholders. It does not validate production ML engineering.

Pair it with a documented prototype or evaluation plan. Show how you selected a model, grounded answers, measured quality, protected data, and defined human oversight. That evidence makes the foundational certification more useful to employers.

Exam-day notes

• You have 90 minutes for 65 total questions.
• Fifty questions are scored and 15 unscored; AWS does not identify them.
• Question types include multiple choice, multiple response, ordering, and matching.
• The US fee is $100 before tax or regional adjustments.
• There is no penalty for guessing, so answer every item.

Prep resources

← More Amazon Web Services exams