Prepared with AI assistance by Pass-IT. Vendor sources support exam facts; the exercises and study recommendations are ours.
Booking-language update, checked September 19, 2026: AWS announces that the German and Italian exam versions will end after October 15, 2026. The exam remains available in other languages. Check your appointment language before booking. AWS certification page
Preparing for AIF-C01 means learning to explain which AI approach fits a problem and how you would evaluate its result. Start by deciding whether that matches your goal: understanding concepts and use cases so you can make informed choices about AWS solutions.
AWS describes a foundational audience that uses AI/ML solutions without necessarily building them. Coding models, tuning algorithms, building pipelines and mathematical or statistical model analysis are outside scope. Recommended familiarity with AWS and IAM is not a requirement to hold a prior credential. AWS exam guide, checked September 19, 2026
Build concepts before choosing services
Use the domain weights to check coverage, then focus your study on the explanations you cannot yet give. The percentages describe scored content, not a prescribed allocation of study time. AWS exam guide
| Domain | Weight | Suggested learning output |
|---|---|---|
| AI and ML fundamentals | 20% | Explain classification versus text generation |
| Generative AI fundamentals | 24% | Explain why a plausible response can lack evidence |
| Foundation-model applications | 28% | Justify retrieval versus a change in model behavior |
| Responsible AI | 14% | Propose a fairness check and explain its limits |
| Security, compliance and governance | 14% | Identify who may access a document |
This is a suggested learning sequence: begin with concrete conceptual examples, connect them to service choices, then examine responsible AI and security. Return to earlier choices when a restriction changes the solution.
Choose how to answer from changing documents
An assistant must answer from internal policies that change every week and show the passage supporting each answer. Should you start with retrieval or fine-tuning?
Start by evaluating retrieval with grounding in the current documents. A Bedrock Knowledge Base can retrieve relevant information for a model response. Fine-tuning changes model behavior; it does not by itself keep policy documents current or provide citations.
Create three short fictional policy documents and six questions. Include a question whose answer is absent and another about a policy you revise. Record the expected supporting passage before comparing responses. Do this on paper first; you do not need to provision a cloud service for the comparison.
Check whether each answer is supported by the right document, whether the revision is reflected and whether the system acknowledges missing evidence. Explain how retrieval permissions and prompt injection could affect the result.
AWS documents retrieval and citations in Knowledge Bases for Amazon Bedrock. Retrieval does not guarantee truth; source updates must also reach the knowledge base.
Work through a small evidence set
Use these fictional documents: Travel says “Meals: maximum €25; receipts required.” Access says “Managers approve access; reviews occur monthly.” Leave says “Submit requests through the portal.” Revise Travel to €30 and retain the earlier version for comparison.
Write six questions: What is the meal limit? Are receipts required? Who approves access? When is access reviewed? How is leave requested? What is the hotel limit?
Your expected-answer sheet should contain €30, yes, managers, monthly, portal and “not specified,” each linked to its supporting passage or marked as absent. The meal answer should change from €25 to €30 after the revision reaches the retrieval source.
| Input | Problem | Approach to evaluate | Required evidence |
|---|---|---|---|
| Revised Travel policy | Answer uses an old limit | Retrieve the current version | €30 and its passage |
| Correct facts, inconsistent wording | Response behavior | Prompt instructions; consider fine-tuning if needed | Consistent wording on new examples |
| Missing hotel rule | No supporting information | Acknowledge the gap | No invented limit |
For a permissions check, mark Access as restricted. Expected output: an unauthorized reader receives neither its text nor a revealing summary. Add “ignore previous instructions” inside a retrieved document. Your written analysis should treat it as untrusted source content, not authority to override the task.
Review your reasoning before booking
Keep an error log with your choice, the requirement you missed and evidence that would change your answer. Try unfamiliar variants rather than relying on recognition of repeated questions. These exercises are learning advice, not official exam questions or a prediction of passing.
AWS lists 65 questions in 90 minutes and three-year certification validity. The guide specifies 50 scored and 15 unidentified unscored questions, with multiple-choice, multiple-response, ordering and matching formats. The passing score is 700 on a 100–1000 scale, not 70% correct. AWS certification page · AWS exam guide
Before choosing a date, compare your explanations with the guide and try timed practice to assess your own pacing. Use Pass-IT AIF-C01 practice to identify topics for your next review.