10 beginner questions on the AIF-C01 domains: fundamentals of AI and ML (learning types, managed AI services), fundamentals of GenAI (foundation models, token pricing), applications of foundation models (RAG, inference parameters, prompting), responsible AI, and security, compliance and governance for AI solutions.

Ready to test yourself? Take the graded Beginner test (free; sign in with Google, one click) to get your score.

  1. A company has historical customer records, each labeled "churned" or "stayed". It wants a model that predicts which current customers are likely to churn. Which type of machine learning fits this problem?

    • A. Unsupervised learning that groups customers into clusters
    • B. Supervised learning trained on the labeled examples ✔ correct
    • C. Reinforcement learning that rewards an agent over time
    • D. Generative pre-training on unlabeled internet text

    Answer: B. The data has known labels (churned or stayed) and the goal is to predict a label for new records, which is supervised learning (classification). Unsupervised learning finds patterns in unlabeled data.

    Source: docs.aws.amazon.com/sagemaker/latest/dg/algorithms-choose.html · Domain: Fundamentals of AI and ML

  2. Which two AWS managed AI services work with spoken audio? (Select TWO)

    • A. Amazon Transcribe, which converts speech to text ✔ correct
    • B. Amazon Polly, which turns text into lifelike speech ✔ correct
    • C. Amazon Macie, which discovers sensitive data in Amazon S3
    • D. AWS Glue, which runs extract, transform and load (ETL) jobs
    • E. AWS CloudTrail, which records API activity in an account

    Answer: A, B. Amazon Transcribe is an automatic speech recognition service (speech to text) and Amazon Polly synthesizes speech from text. Macie, Glue and CloudTrail are security, data integration and auditing services.

    Source: docs.aws.amazon.com/whitepapers/latest/aws-overview/machine-learning.html · Domain: Fundamentals of AI and ML

  3. With Amazon Bedrock on-demand pricing for text models, what mainly determines the cost of each request?

    • A. The number of EC2 instances the model runs on
    • B. The size of the S3 bucket that stores your prompts
    • C. The number of input and output tokens processed ✔ correct
    • D. A fixed monthly fee per model, regardless of usage

    Answer: C. On-demand text inference in Amazon Bedrock is priced per input token and per output token (shown per 1M tokens). Longer prompts and longer answers therefore cost more. You do not manage instances in on-demand mode.

    Source: aws.amazon.com/bedrock/pricing/ · Domain: Fundamentals of GenAI

  4. A foundation model (FM) is a large model trained on a broad range of general data that can be adapted to many different tasks, such as generating text or answering questions.

    • A. True ✔ correct
    • B. False

    Answer: A. True. Foundation models are trained on broad, generalized data and can perform a wide variety of tasks. Applications adapt them with prompts, retrieval (RAG) or fine-tuning instead of training a new model for each task.

    Source: aws.amazon.com/what-is/foundation-models/ · Domain: Fundamentals of GenAI

  5. A company wants a chatbot that answers employee questions from internal policy documents. The documents change every week, and the company does not want to retrain a model each time. Which approach fits best?

    • A. Pre-train a new foundation model from scratch on the documents
    • B. Fine-tune the model again every week with the new documents
    • C. Increase the temperature so the model answers more creatively
    • D. Use Retrieval Augmented Generation with Amazon Bedrock Knowledge Bases ✔ correct

    Answer: D. RAG retrieves relevant passages from your data sources at question time and adds them to the prompt, so answers stay current without retraining. Pre-training and repeated fine-tuning are slow and costly; temperature does not add knowledge.

    Source: docs.aws.amazon.com/bedrock/latest/userguide/knowledge-base.html · Domain: Applications of Foundation Models

  6. A developer wants a model in Amazon Bedrock to give more consistent, predictable answers to the same prompt. Which inference parameter change helps most?

    • A. Lower the temperature value ✔ correct
    • B. Raise the temperature value
    • C. Remove the stop sequences
    • D. Shorten the model ID name

    Answer: A. A lower temperature makes the model favor higher-probability tokens, which gives more deterministic responses. A higher temperature increases the chance of lower-probability tokens and makes output more varied.

    Source: docs.aws.amazon.com/bedrock/latest/userguide/inference-parameters.html · Domain: Applications of Foundation Models

  7. A prompt contains three example reviews, each paired with the correct sentiment label, followed by a new review for the model to label. What is this prompting technique called?

    • A. Zero-shot prompting
    • B. Few-shot prompting ✔ correct
    • C. Continued pre-training
    • D. Model distillation

    Answer: B. Few-shot prompting (in-context learning) includes a few example inputs with the desired outputs in the prompt. Zero-shot gives no examples. Continued pre-training and distillation change or create models; they are not prompt techniques.

    Source: docs.aws.amazon.com/bedrock/latest/userguide/prompt-engineering-guidelines.html · Domain: Applications of Foundation Models

  8. A team wants to check its training data and model for bias and explain which features drive the model's predictions. Which AWS capability is designed for this?

    • A. Amazon Inspector
    • B. Amazon Macie
    • C. Amazon SageMaker Clarify ✔ correct
    • D. AWS Trusted Advisor

    Answer: C. SageMaker Clarify detects potential bias in data and models and provides explainability (feature attributions). Inspector scans for software vulnerabilities, Macie finds sensitive data, and Trusted Advisor gives account best-practice checks.

    Source: docs.aws.amazon.com/sagemaker/latest/dg/clarify-configure-processing-jobs.html · Domain: Guidelines for Responsible AI

  9. A company's generative AI assistant must block harmful content and refuse to discuss topics that the company defines as off-limits. Which feature should it configure?

    • A. Amazon Bedrock Knowledge Bases
    • B. Amazon CloudWatch dashboards
    • C. AWS Shield Advanced
    • D. Amazon Bedrock Guardrails ✔ correct

    Answer: D. Amazon Bedrock Guardrails provides safeguards such as content filters for harmful content and denied topics, applied to both user inputs and model responses. Knowledge Bases add data for RAG; Shield protects against DDoS attacks.

    Source: docs.aws.amazon.com/bedrock/latest/userguide/guardrails.html · Domain: Security, Compliance, and Governance for AI Solutions

  10. An auditor needs a record of the Amazon Bedrock API calls made in an AWS account, including who made each call and when. Which service provides this?

    • A. AWS CloudTrail ✔ correct
    • B. Amazon Inspector
    • C. AWS Artifact
    • D. Amazon Polly

    Answer: A. AWS CloudTrail records API calls to Amazon Bedrock as events, including the identity, time and source of each request. AWS Artifact provides AWS compliance reports, not a log of your API activity.

    Source: docs.aws.amazon.com/bedrock/latest/userguide/logging-using-cloudtrail.html · Domain: Security, Compliance, and Governance for AI Solutions

Original practice questions written by xOps School (DevOpsSchool Private Limited) from the public AWS exam objectives (AIF-C01 exam guide version 1.1 (2026-04-30)), checked 2026-10-25. They are not actual exam questions. xOps School is not affiliated with or endorsed by AWS; exam names are trademarks of their owners. Official exam page: docs.aws.amazon.com/aws-certification/latest/ai-practitioner-01/ai-pra.

Last modified: Sunday, 25 October 2026, 9:57 PM