PasswellLog inStart free
AWS

AWS AI Practitioner prep

For business, product and technical people who use or buy AI on AWS and want to prove they understand machine learning, generative AI, agents and responsible, secure AI.

Start freeUnlock everything · $19

The real exam

Exam codeAIF-C01 (exam guide v1.2, 29 Oct 2026, with the agentic AI content added in v1.1)
Questions65 (50 scored, 15 unscored and unidentified)
Time90 minutes
Pass mark700 on a 100 to 1000 scaled score
Question typesMultiple choice, multiple response, ordering (put 3 to 5 items in order) and matching (pair prompts with answers)
ScoringCompensatory: you pass on the overall score, not per domain. Multiple response, ordering and matching are all or nothing. No penalty for guessing
FeeUSD 100
DeliveryPearson VUE test centre or online proctored
ResultsProvisional pass or fail on screen, scaled score within 5 business days
RetakeWait 14 days after a fail, full fee each time
Valid for3 years
LevelFoundational: no coding, maths, feature engineering or hyperparameter tuning

What you get

32
lessons
4
hours of reading
170
original questions
2
full mock exams

Mocks follow the real rules: 65 questions in 90 minutes, flag and review.

Syllabus

Fundamentals of AI and ML20%

AI, ML, deep learning and agentic AI terms, learning and data types, inference options, use cases, the ML lifecycle, MLOps and metrics.

Fundamentals of Generative AI24%

Tokens, embeddings, transformers, token pricing, context engineering, agents, MCP, limits of GenAI and the AWS GenAI stack incl. Bedrock, AgentCore, Strands Agents, Kiro and Amazon Quick.

Applications of Foundation Models28%

Model selection, inference parameters, RAG and vector stores, prompt engineering, fine-tuning and distillation, and evaluation with ROUGE, BLEU, BERTScore and LLM-as-a-judge.

Guidelines for Responsible AI14%

Bias, fairness, robustness and veracity, Bedrock Guardrails, legal risks, dataset quality, transparency, explainability and Model Cards.

Security, Compliance, and Governance for AI Solutions14%

IAM, encryption, PrivateLink, Macie, AgentCore Identity and Policy, prompt injection, grounding, audit logging, governance services and the GenAI Security Scoping Matrix.

Course outline

  1. AI and ML foundations · 5 lessons
  2. The ML lifecycle, MLOps and metrics · 3 lessons
  3. Generative AI fundamentals · 4 lessons
  4. Agentic AI · 5 lessons
  5. Building applications with Amazon Bedrock · 5 lessons
  6. Customising and evaluating foundation models · 4 lessons
  7. Responsible AI · 3 lessons
  8. Security, compliance and governance for AI · 3 lessons

Try a question

A warehouse robotics team trains software to move a robot arm. The software tries actions in a simulator, gets points for placing parcels safely and loses points for drops. Over millions of attempts it improves. Which learning approach is this?
A. Unsupervised clustering that groups similar arm movements together by their sensor readings
B. Transfer learning, where a model trained on one task is reused as the starting point for another
C. Reinforcement learning, where an agent learns from rewards and penalties
D. Supervised learning with a labelled dataset of correct arm movements prepared by engineers
Show answer
Answer: C. Reinforcement learning has an agent, an environment, actions and rewards. It suits robotics, games and control problems. RLHF (reinforcement learning from human feedback) uses the same idea with human preference scores to align language models.
A model scores 99% accuracy on its training data but only 61% on new data it has not seen. What is the MOST likely problem?
A. Overfitting, because the model memorised the training data
B. Data drift, because production traffic has slowly changed since the model was first deployed
C. Underfitting, because the model is too simple to learn the patterns in the training data
D. Label leakage, because the test data contains the target value as an extra input column
Show answer
Answer: A. Overfitting = high variance: the model fits noise in the training set. Fixes include more and more varied data, simpler models, regularisation and early stopping. Underfitting = high bias: poor on both training and new data.

Passwell is not affiliated with Amazon Web Services. AWS, Amazon Bedrock, Amazon SageMaker AI and related names are trademarks of Amazon.com, Inc. or its affiliates. All questions and lessons are original practice material. Facts checked 2026-11-14.