Which statement best describes artificial intelligence in a computing system?

It enables software to perform tasks that normally require human-like perception, reasoning, or decision-making.

A team changes a model setting to make responses more varied and creative. Which GenAI concept is most closely related to that behavior?

Temperature or sampling configuration

A company wants a system that reads incoming support emails, determines whether a refund is likely appropriate, drafts a response, checks the order system, and opens a manager approval task when the refund exceeds a threshold. Which design best matches the requirement?

Combine classification, generative drafting, and an agentic workflow with controlled tool access and approval steps.

A team uses the same model for brainstorming taglines and drafting regulated notices. How should inference parameters generally differ?

Brainstorming may use more randomness, while regulated notices usually need more constrained settings.

Which task is a common use case for a generative AI model?

Summarizing a long policy document into key points for review.

A company processes insurance claims and wants a model to summarize adjuster notes, repair estimates, and customer emails. Which selection approach is most defensible?

Test candidate models on representative claims with criteria for accuracy, omissions, citations, latency, and cost.

How does transparency differ from explainability in an AI system?

Transparency shares relevant system information, while explainability helps interpret specific outputs.

What does retrieval-augmented generation add to a foundation model application?

It retrieves relevant external information and provides it as context before generation.

A company deploys a Bedrock-based assistant for employee benefits questions. Users report that it sometimes gives medical advice and sometimes refuses harmless policy questions. What should the team do?

Tune guardrails with traces, denied topics, and false-positive and false-negative tests.

A company wants its VPC-hosted application to call supported AWS AI services without sending traffic over the public internet. Which feature is most relevant?

VPC endpoints or AWS PrivateLink

AWS Certified AI Practitioner (AIF-C01) Practice Test: 10 Free Questions (2026)

Exam-specific sample

Questions, answers, and option-by-option explanations

Each explanation identifies why the correct option fits and why the distractors do not.

Easy Fundamentals of AI and ML

Question 1

Which statement best describes artificial intelligence in a computing system?
1. It stores business data in a managed relational database for transactional workloads.
2. It enables software to perform tasks that normally require human-like perception, reasoning, or decision-making.
3. It replaces all human review by claiming to ensure correct decisions from any input data.
4. It compresses files and transfers them between systems with lower network latency.
Correct answer: B. It enables software to perform tasks that normally require human-like perception, reasoning, or decision-making.
- A. Incorrect. A relational database can support an application, but storing transactional data is not what makes a system intelligent.
- B. Correct. AI systems are designed to perform tasks associated with human intelligence, such as recognizing patterns, interpreting language, or making decisions.
- C. Incorrect. AI can assist decisions, but it does not help ensure correctness or reduce the need for human oversight in all cases.
- D. Incorrect. Compression and data transfer are infrastructure functions, not defining characteristics of AI.
Medium Fundamentals of generative AI

Question 2

A team changes a model setting to make responses more varied and creative. Which GenAI concept is most closely related to that behavior?
1. Context window size only
2. Object storage lifecycle policy
3. Temperature or sampling configuration
4. Database primary key selection
Correct answer: C. Temperature or sampling configuration
- A. Incorrect. Context window size affects how much text can be considered, not primarily creativity.
- B. Incorrect. Storage lifecycle policies do not control model response variety.
- C. Correct. Temperature and sampling settings influence randomness and variety in generated outputs.
- D. Incorrect. Primary keys are database design concepts.
Hard Fundamentals of AI and ML

Question 3

A company wants a system that reads incoming support emails, determines whether a refund is likely appropriate, drafts a response, checks the order system, and opens a manager approval task when the refund exceeds a threshold. Which design best matches the requirement?
1. Combine classification, generative drafting, and an agentic workflow with controlled tool access and approval steps.
2. Use only a static keyword rule because the task contains the word refund.
3. Use only image generation because the system needs to create a new artifact.
4. Use a single database index because all steps are data retrieval problems.
Correct answer: A. Combine classification, generative drafting, and an agentic workflow with controlled tool access and approval steps.
- A. Correct. The workflow includes prediction, text generation, tool use, and business approvals, so it needs multiple AI patterns with governance.
- B. Incorrect. A keyword rule is too brittle for the decision, drafting, tool-checking, and approval workflow described.
- C. Incorrect. The new artifact is text, not an image, and the workflow also requires decisions and tool use.
- D. Incorrect. Database indexing may support lookups but cannot perform classification, drafting, or approval orchestration.
Medium Applications of foundation models

Question 4

A team uses the same model for brainstorming taglines and drafting regulated notices. How should inference parameters generally differ?
1. Both tasks should use the same high temperature so style remains consistent.
2. Regulated notices should remove output limits so the model can include optional commentary.
3. Brainstorming should use the most constrained sampling settings to increase novelty.
4. Brainstorming may use more randomness, while regulated notices usually need more constrained settings.
Correct answer: D. Brainstorming may use more randomness, while regulated notices usually need more constrained settings.
- A. Incorrect. High temperature can harm consistency for regulated text.
- B. Incorrect. Output limits can help control notice format, cost, and unapproved additions.
- C. Incorrect. Very constrained sampling usually reduces creative diversity.
- D. Correct. Creative and regulated workflows have different tolerance for variation.
Easy Fundamentals of generative AI

Question 5

Which task is a common use case for a generative AI model?
1. Adding two invoice line items with a fixed formula.
2. Summarizing a long policy document into key points for review.
3. Assigning a static IP address from a network plan.
4. Renaming a file with yesterday's date.
Correct answer: B. Summarizing a long policy document into key points for review.
- A. Incorrect. Fixed arithmetic should use deterministic code.
- B. Correct. Summarization is a common GenAI use case because the model generates concise text from longer content.
- C. Incorrect. Static IP assignment is network configuration, not content generation.
- D. Incorrect. File renaming is simple automation.
Hard Applications of foundation models

Question 6

A company processes insurance claims and wants a model to summarize adjuster notes, repair estimates, and customer emails. Which selection approach is most defensible?
1. Choose the model with the most creative marketing examples.
2. Test candidate models on representative claims with criteria for accuracy, omissions, citations, latency, and cost.
3. Select any model that writes fluent paragraphs about insurance generally.
4. Use only a general public benchmark score with no claims data evaluation.
Correct answer: B. Test candidate models on representative claims with criteria for accuracy, omissions, citations, latency, and cost.
- A. Incorrect. Creative marketing examples do not prove claims-review quality.
- B. Correct. The model should be evaluated on realistic claim materials and workflow requirements.
- C. Incorrect. Fluency does not prove accuracy or completeness.
- D. Incorrect. Public benchmarks do not replace domain-specific testing.
Medium Guidelines for responsible AI

Question 7

How does transparency differ from explainability in an AI system?
1. Transparency is a latency metric, while explainability is a cost metric.
2. Transparency shares relevant system information, while explainability helps interpret specific outputs.
3. Transparency requires disclosing private training records to every user.
4. Transparency and explainability are documentation tasks unrelated to responsible AI.
Correct answer: B. Transparency shares relevant system information, while explainability helps interpret specific outputs.
- A. Incorrect. Neither term means latency or cost measurement.
- B. Correct. Transparency concerns clear disclosure and documentation; explainability concerns understanding model behavior or decisions.
- C. Incorrect. Transparency should not violate privacy or security.
- D. Incorrect. Both are responsible AI features, not unrelated paperwork.
Easy Applications of foundation models

Question 8

What does retrieval-augmented generation add to a foundation model application?
1. It fine-tunes the model weights every time a user asks a question.
2. It retrieves relevant external information and provides it as context before generation.
3. It relies only on the model's pretraining for private company knowledge.
4. It replaces source-document management with a longer system prompt.
Correct answer: B. It retrieves relevant external information and provides it as context before generation.
- A. Incorrect. RAG supplies context at inference time rather than retraining model weights for each question.
- B. Correct. RAG combines retrieval from a knowledge source with model generation.
- C. Incorrect. RAG is used when the answer should come from external or private sources, not only pretraining.
- D. Incorrect. A longer system prompt does not replace managing and retrieving source content.
Hard Guidelines for responsible AI

Question 9

A company deploys a Bedrock-based assistant for employee benefits questions. Users report that it sometimes gives medical advice and sometimes refuses harmless policy questions. What should the team do?
1. Tune guardrails with traces, denied topics, and false-positive and false-negative tests.
2. Remove the guardrails so harmless benefits questions are less likely to be blocked.
3. Increase the response token limit before checking whether refusals are accurate.
4. Judge the assistant by average latency rather than safe and correct behavior.
Correct answer: A. Tune guardrails with traces, denied topics, and false-positive and false-negative tests.
- A. Correct. The issue involves both missed risky content and overblocking, so guardrail evaluation and tuning are needed.
- B. Incorrect. Removing controls would increase unsafe medical-advice risk.
- C. Incorrect. Token limits do not solve policy misclassification.
- D. Incorrect. Latency does not measure safe and correct behavior.
Medium Security, compliance, and governance for AI solutions

Question 10

A company wants its VPC-hosted application to call supported AWS AI services without sending traffic over the public internet. Which feature is most relevant?
1. VPC endpoints or AWS PrivateLink
2. Public S3 website hosting
3. Route 53 domain registration
4. CloudFront signed cookies only
Correct answer: A. VPC endpoints or AWS PrivateLink
- A. Correct. VPC endpoints and PrivateLink can provide private connectivity to supported AWS services.
- B. Incorrect. Public website hosting exposes content rather than private service access.
- C. Incorrect. Domain registration does not create private API connectivity.
- D. Incorrect. Signed cookies control CloudFront access, not private AI service calls.

More

AWS Certified AI Practitioner (AIF-C01) Practice Test: 10 Free Questions

Questions, answers, and option-by-option explanations

Question 1

Question 2

Question 3

Question 4

Question 5

Question 6

Question 7

Question 8

Question 9

Question 10

Continue with the full exam-prep cluster

How can I help?