Skip to content

Free AWS Certified AI Practitioner (AIF-C01) practice test

The AWS Certified AI Practitioner (AIF-C01) validates a foundational understanding of AI, machine learning, and generative AI on AWS: what the technologies are, when to apply them, and how to use them responsibly and securely. This site gives you 321 original practice questions plus 16 interactive items in the exam's own formats, free and without an account.

AIF-C01 is not just multiple choice: the real exam also uses multiple-response ("Select TWO"), ordering, and matching questions. Our simulator reproduces all four formats, runs 90 timed minutes, and scores on the real 100 to 1,000 scale with the 700 pass line. Every topic links into the free 15-hour freeCodeCamp course at the matching chapter.

Start the free AIF-C01 practice test

What AIF-C01 covers

AIF-C01 domainExam weightQuestions here
Fundamentals of AI and ML20%about 64
Fundamentals of Generative AI24%about 77
Applications of Foundation Models28%about 90
Guidelines for Responsible AI14%about 45
Security, Compliance, and Governance for AI14%about 45

Real exam facts: 65 questions (50 scored, 15 unscored), 90 minutes, scaled 100 to 1,000, passing score 700.

Practice questions from this bank

Sample question 1 · AIF-C01 · Fundamentals of AI and ML

Which statement best describes the relationship among artificial intelligence, machine learning, and deep learning?

  1. ADeep learning is a subset of machine learning, which is itself a subset of artificial intelligence
  2. BMachine learning and deep learning are separate fields that both contain artificial intelligence
  3. CArtificial intelligence is a subset of deep learning that focuses on neural networks
  4. DMachine learning replaces artificial intelligence, while deep learning replaces machine learning
Show answer and explanation

A. Deep learning is a subset of machine learning, which is itself a subset of artificial intelligence

AI is the broadest field, ML is the subset of AI that learns patterns from data, and deep learning is the subset of ML built on multi-layer neural networks. The other orderings invert this hierarchy or wrongly treat the three as rival, unrelated fields.

Topic: freeCodeCamp: AI, ML, and Deep Learning Basics

Sample question 2 · AIF-C01 · Fundamentals of Generative AI

A data science team is explaining to executives how a new chatbot project differs from the fraud-scoring models the company already runs in production. Which capability distinguishes the chatbot's underlying model from the existing predictive models?

  1. AIt requires labeled training data, while predictive models can learn without labels
  2. BIt classifies inputs into fixed categories more accurately than older models
  3. CIt creates new content, such as original text, rather than assigning labels or scores to inputs
  4. DIt always produces identical outputs for identical inputs
Show answer and explanation

C. It creates new content, such as original text, rather than assigning labels or scores to inputs

Generative AI learns the underlying distribution of its training data so it can produce novel content, while predictive models output labels or scores. The labeled-data option is backwards since foundation models pretrain largely on unlabeled data, and sampling makes generative output non-deterministic rather than identical.

Topic: freeCodeCamp: Generative AI Concepts

Sample question 3 · AIF-C01 · Applications of Foundation Models

Which statement best describes what a RAG pipeline adds to a foundation model?

  1. AIt permanently updates the model weights with domain knowledge
  2. BIt looks up cached answers from earlier sessions so repeated prompts skip the model
  3. CIt retrieves related information from an external source and passes it to the model
  4. DIt screens generated responses for harmful or restricted content
Show answer and explanation

C. It retrieves related information from an external source and passes it to the model

RAG fetches relevant content from an external knowledge source at inference time and injects it into the model's input; the weights are never touched. Looking up cached answers to repeated prompts describes response caching, which reuses old output instead of grounding a new one, and permanently updating weights describes fine-tuning.

Topic: freeCodeCamp: Retrieval-Augmented Generation

Sample question 4 · AIF-C01 · Guidelines for Responsible AI

A loan-approval model approves urban applicants at a much higher rate than rural applicants who have similar financial profiles. After reviewing the training pipeline, what is the most likely cause?

  1. AThe feature pipeline served stale records for some applicants at inference time
  2. BThe temperature parameter was set too high during scoring
  3. COne applicant population was heavily underrepresented in the training records
  4. DThe model was trained for too many epochs and memorized noise
Show answer and explanation

C. One applicant population was heavily underrepresented in the training records

Skewed representation in training data is the classic source of group-level performance gaps: the model saw too few examples from one population to learn its patterns. Temperature controls randomness in generative output and cannot produce a systematic disparity between demographic groups.

Topic: freeCodeCamp: Bias, Fairness, and Dataset Balance

Sample question 5 · AIF-C01 · Security, Compliance, and Governance for AI

A developer is deploying a customer-facing application on Amazon EC2 that calls a single foundation model through Amazon Bedrock. Which approach follows the principle of least privilege for the application's access?

  1. AStore an administrator's access keys in the application's configuration file and rotate them monthly
  2. BGrant the instance full Bedrock permissions and review CloudTrail weekly for unexpected activity
  3. CAttach an IAM role to the instance that allows the invoke action only on the specific model's ARN
  4. DUse the account root user credentials with multi-factor authentication enabled
Show answer and explanation

C. Attach an IAM role to the instance that allows the invoke action only on the specific model's ARN

An instance role scoped to a single action on a single model ARN grants exactly what the workload needs and supplies temporary, automatically rotated credentials. Broad permissions with after-the-fact log review, stored admin keys, and root credentials all violate least privilege even when paired with rotation or MFA.

Topic: freeCodeCamp: Securing AI Systems on AWS

Sample question 6 · AIF-C01 · Fundamentals of AI and ML

Developers are training an agent to navigate a simulated warehouse: it earns positive feedback for reaching the loading dock quickly and penalties for collisions, improving over repeated trials. Which learning paradigm is being used?

  1. ASupervised learning
  2. BUnsupervised learning
  3. CReinforcement learning
  4. DTransfer learning
Show answer and explanation

C. Reinforcement learning

An agent that improves by acting in an environment and receiving rewards and penalties is the hallmark of reinforcement learning. Supervised learning would need labeled examples of correct moves, and transfer learning describes reusing knowledge from a previously trained model, not this trial-and-error training loop.

Topic: freeCodeCamp: Supervised, Unsupervised, and Reinforcement Learning

Sample question 7 · AIF-C01 · Fundamentals of Generative AI

A financial analyst pastes a 400-page annual report into a chat assistant and asks for a summary, but the model responds as if it only saw the final chapters. What is the most likely cause?

  1. AThe document exceeded the model's context window, so the earlier text never reached the model
  2. BThe temperature setting was too high for long documents
  3. CThe embeddings for the earlier chapters expired from the vector store
  4. DA guardrail flagged the earlier chapters as a denied topic
Show answer and explanation

A. The document exceeded the model's context window, so the earlier text never reached the model

A model can only attend to the tokens that fit inside its context window, and oversized input gets truncated. Temperature affects how randomly words are chosen, not how much input the model can hold.

Topic: freeCodeCamp: Tokens, Embeddings, and Vectors

Sample question 8 · AIF-C01 · Applications of Foundation Models

A support organization has 25,000 past tickets, each tagged with its correct routing category, and wants the model to assign categories consistently without depending on long prompts. What should the team do?

  1. APaste representative tickets into every prompt as in-context examples
  2. BStore past tickets in a vector index and retrieve similar ones at query time
  3. CCustomize the model on the ticket-category pairs so the mapping is learned into its weights
  4. DEnable guardrails that restrict outputs to the approved category list
Show answer and explanation

C. Customize the model on the ticket-category pairs so the mapping is learned into its weights

With a large labeled dataset and a need for consistent task behavior, fine-tuning trains the mapping into the model itself. In-context examples consume tokens on every call and can behave inconsistently, and guardrails block or allow content rather than teach a classification skill.

Topic: freeCodeCamp: Fine-Tuning and Continued Pre-Training

Practice all of them free

Common questions

How many questions are on the AIF-C01 exam?

65 questions in 90 minutes: 50 scored plus 15 unscored questions AWS evaluates for future exams. Formats include multiple choice, multiple response, ordering, and matching. Our simulator mirrors all four.

What score do I need to pass?

AWS reports a scaled score from 100 to 1,000 with a minimum passing score of 700. Scoring is compensatory: you pass on the overall exam, not on each section.

Do I need AWS experience first?

AWS recommends up to six months of exposure to AI and ML workloads on AWS. The target candidate uses AI services rather than builds models; no coding or data-science background is required.

Is this actually free?

Yes. Every question, the exam simulator, and progress tracking are free with no account. Topic reviews link to Andrew Brown's free 15-hour AIF-C01 course on freeCodeCamp's YouTube channel.

More free practice tests

Back to all cert900 exams