AWS AIP-C01 Sample Questions with Explanations

These original sample questions are designed to help you check how the exam topics appear in decision-style prompts. They are not taken from the live exam.

Use these sample questions as a guided self-assessment for AWS Certified Generative AI Developer - Professional (AIP-C01) topics such as foundation model integration, Amazon Bedrock application design, retrieval, agents, tool use, guardrails, data protection, evaluation, monitoring, troubleshooting, latency, and cost. The prompts emphasize production design rather than isolated service trivia.

Where these questions fit in the AIP-C01 guide

The sample set below is part of the AWS AIP-C01 guide path:

• AWS certification hub
• AIP-C01 exam guide
• AIP-C01 cheat sheet
• AIP-C01 study plan
• AIP-C01 FAQ
• AIP-C01 resources
• AIP-C01 glossary

AIP-C01 production GenAI sample questions

Work through each prompt before opening the explanation. AIP-C01 questions usually reward answers that ground model output, protect data, control tools, and make behavior measurable.

---

Question 1

Topic: Grounding an internal policy assistant

A company is building an internal assistant that answers HR policy questions from approved documents. Answers must be grounded in current policy, include citations, and avoid using stale documents that are no longer approved. What is the best design?

• A. Fine-tune a model once on the complete HR document archive and tell users to verify answers manually.
• B. Paste the most common HR policy text into the system prompt and increase the maximum output length.
• C. Use a retrieval-based design over an approved knowledge source, keep document metadata current, and require cited grounded responses.
• D. Let the model search the public web for HR answers when internal documents are incomplete.

Best answer: C

Explanation: The requirement is current, approved, cited policy content. Retrieval keeps the source corpus maintainable and lets the application select approved passages at runtime instead of relying on model memory.

Why the other choices are weaker:

• A makes freshness and citation quality difficult because the model internalizes old content.
• B is brittle and cannot scale across policy versions.
• D leaves the trusted internal data boundary and may return irrelevant or unauthorized material.

What this tests: Choosing retrieval-augmented generation for grounded, governed, citation-backed enterprise answers.

Related topics: Retrieval; Amazon Bedrock; Grounding; Governance

---

Question 2

Topic: Controlling an agent action

An agent can summarize customer issues and create refund requests through an internal API. The business allows automatic requests only below a dollar threshold and requires every action to be traceable. Which implementation is strongest?

• A. Give the agent a broad administrator role and let the model decide when to call the refund API.
• B. Disable logging to reduce the amount of customer data stored after each action.
• C. Expose refund creation as a constrained tool with scoped IAM permissions, input validation, threshold checks, idempotency, and audit logging.
• D. Use a larger model so it is less likely to make mistakes when creating refunds.

Best answer: C

Explanation: A production agent should not receive broad authority. The action path needs least privilege, validated inputs, business rules, duplicate protection, and logs that show what happened.

Why the other choices are weaker:

• A creates excessive privilege and weak control around business actions.
• B weakens investigation and compliance for a financial workflow.
• D may improve language reasoning, but it does not enforce permissions or policy limits.

What this tests: Securing agent tool use with IAM boundaries, validation, policy checks, and traceability.

Related topics: Agents; Tool use; IAM; Audit logging

---

Question 3

Topic: Troubleshooting poor generated answers

After a new release, a support chatbot starts giving vague answers even though the model endpoint is healthy. Logs show retrieval returns only one short passage for many questions, and user satisfaction drops. What should the team investigate first?

• A. Only the model size, because vague answers always mean the model is too small.
• B. The color and placement of the chat widget in the application.
• C. The monthly cloud bill, because answer quality is primarily a cost-allocation problem.
• D. The retrieval pipeline, including chunking, embeddings, filters, search quality, and prompt context assembly.

Best answer: D

Explanation: The logs point to a retrieval problem. If the model receives thin or poorly selected context, the generated answer may be vague even when the model endpoint is functioning.

Why the other choices are weaker:

• A ignores the observed retrieval signal and jumps to an expensive model change.
• B may affect user experience, but it does not explain missing context.
• C matters operationally, but it is not the first quality root cause in this scenario.

What this tests: Troubleshooting GenAI quality by reading telemetry and separating model behavior from retrieval behavior.

Related topics: RAG troubleshooting; Embeddings; Prompt context; Monitoring

---

Question 4

Topic: Protecting sensitive source data

A legal department wants a GenAI app to summarize confidential contracts. The application must keep source files encrypted, limit access to authorized users, prevent broad model access to unrelated documents, and retain logs for review. Which design best fits?

• A. Apply encryption, least-privilege IAM, document-level access filters, network controls where appropriate, and centralized audit logs.
• B. Put all contracts into one public object bucket and rely on prompt wording to prevent disclosure.
• C. Remove all logs so sensitive prompts cannot be reviewed by anyone.
• D. Allow users to upload documents into any account as long as the model returns a summary.

Best answer: A

Explanation: A production GenAI architecture still needs ordinary cloud security controls. Encryption, least privilege, filtering, network boundaries, and audit logs protect source data and make behavior reviewable.

Why the other choices are weaker:

• B exposes confidential data and treats the prompt as a security boundary.
• C removes reviewability and may conflict with audit requirements.
• D creates uncontrolled data placement and access risk.

What this tests: Applying data protection, access control, and auditability to GenAI source material and outputs.

Related topics: Data protection; Least privilege; Audit logs; Confidential data

Independent study note

Tech Exam Lexicon and IT Mastery are independent study tools. They are not affiliated with, endorsed by, or sponsored by AWS or any certification body.