AIF-C01 Practice Test 22 — 15 Questions

Question 1

A company is building a generative AI application using Amazon Bedrock. They need to ensure that the model does not generate responses containing personally identifiable information (PII) such as credit card numbers or social security numbers. Which Bedrock feature should they configure?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Amazon Bedrock Guardrails. Amazon Bedrock Guardrails is the correct feature because it provides configurable safeguards to filter and block sensitive content, including PII like credit card numbers and social security numbers, from being generated in model responses. It allows you to define denied topics, content filters, and sensitive information filters that are applied at inference time, ensuring compliance with data privacy requirements without modifying the underlying foundation model.

Answer

Amazon Bedrock Knowledge Bases

Answer

Amazon Bedrock Agents

Answer

Amazon Bedrock batch inference

Question 2

A company wants to ensure that only approved machine learning models are deployed to production on Amazon SageMaker. Which combination of services can enforce this governance requirement?

Accepted Answer

AWS Config custom rules and AWS IAM policies. AWS Config custom rules can evaluate SageMaker model deployment configurations against defined policies (e.g., requiring models to be from an approved registry), and AWS IAM policies can restrict who can create or update endpoints, together enforcing that only approved ML models are deployed. This combination provides both continuous compliance checking and access control, directly addressing the governance requirement.

Answer

AWS CodePipeline and Amazon CodeGuru

Answer

Amazon CloudWatch Events and AWS CloudTrail

Answer

AWS Organizations and AWS Artifact

Question 3

A data scientist uses Amazon Bedrock. The model responses are too long. Which parameter should they adjust to limit the output length?

Accepted Answer

max_tokens. The `max_tokens` parameter directly controls the maximum number of tokens (words or subwords) the model can generate in a single response. By reducing this value, the data scientist caps the output length, preventing overly long responses. Temperature and top_p affect randomness and diversity, not length, while stop sequences define when generation halts but do not enforce a hard token limit.

Answer

temperature

Answer

stop sequences

Answer

top_p

Question 4

A company wants to personalize product recommendations for its e-commerce website. The recommendation engine should adapt to each user's browsing and purchase history in real time. Which AWS service is MOST suitable?

Accepted Answer

Amazon Personalize. Amazon Personalize is the correct choice because it is a fully managed ML service specifically designed to build real-time personalized recommendation systems. It uses the same technology as Amazon's own recommendation engine, processing user-item interaction data (browsing and purchase history) to generate tailored product suggestions with sub-second latency via a real-time inference endpoint.

Answer

Amazon Rekognition

Answer

Amazon Comprehend

Answer

Amazon Forecast

Question 5

A marketing team is using a foundation model to generate marketing copy. Which THREE of the following should they consider to ensure responsible and cost-effective use?

Accepted Answer

Bias mitigation to avoid unfair stereotypes. Option A (Bias mitigation to avoid unfair stereotypes) is correct because foundation models can reproduce or amplify biases present in their training data, and marketing copy that relies on unfair stereotypes creates reputational, legal, and ethical risk, so teams should apply bias detection and mitigation techniques. Option B (Cost per token for the model) is correct because foundation models are typically billed by input and output tokens, so tracking cost per token directly supports cost-effective use and lets the team choose the most economical model or prompt design for the workload. Option D (Toxicity detection in generated content) is correct because generative models can produce offensive, harmful, or brand-damaging text, so toxicity detection and filtering are needed to keep marketing content responsible and safe to publish. Option C (Model size, number of parameters) is not one of the required answers here because parameter count is only an indirect proxy for capability and cost, and by itself it does not ensure responsible or cost-effective use. Option E (Latency of model inference) is not one of the required answers because inference latency affects user experience and responsiveness, not the responsible-use or token-cost concerns the scenario asks about.

Answer

Model size (number of parameters)

Answer

Latency of model inference

Question 6

Refer to the exhibit. A data scientist created this endpoint config for a foundation model in Amazon SageMaker. However, the endpoint fails to scale under load. What is the most likely reason?

Accepted Answer

Missing AutoScaling configuration. The endpoint fails to scale under load because the endpoint configuration shown lacks an AutoScaling policy. Without AutoScaling, SageMaker will not automatically adjust the number of instances based on traffic, so even if the initial instance count is 1, the endpoint cannot add more instances to handle increased load. AutoScaling must be explicitly configured via Application Auto Scaling to define scaling policies and target tracking metrics.

Answer

Variant weight is 1.0

Answer

Instance type is too small

Answer

InitialInstanceCount is 1

Question 7

A company uses Amazon Bedrock to build a question-answering system. Which THREE features of Amazon Bedrock can improve answer accuracy? (Choose three.)

Accepted Answer

Retrieval Augmented Generation (RAG). Retrieval Augmented Generation (RAG) (A) is correct because it grounds the model's responses in an external knowledge base retrieved from sources like Amazon OpenSearch Serverless or Aurora, injecting relevant, up-to-date context into the prompt so answers are factually accurate rather than hallucinated. Model fine-tuning (C) is correct because adapting a foundation model on domain-specific labeled data adjusts its weights to better match the company's terminology, style, and task patterns, directly raising answer quality for that use case. Prompt engineering (E) is correct because carefully designing instructions, few-shot examples, and output formatting in the prompt steers the model toward more precise, relevant, and consistent answers without retraining. Auto-scaling of provisioned throughput (B) only affects performance and cost under load, not answer correctness, and encryption at rest (D) is a security control protecting stored data, neither of which improves the accuracy of generated answers.

Answer

Auto-scaling of provisioned throughput

Answer

Encryption at rest

Question 8

Which TWO actions can help mitigate bias in a face recognition model trained on AWS? (Select two.)

Accepted Answer

Ensure the training dataset is balanced across demographics. Options A and B are correct. Mitigating bias in a face recognition model requires a balanced training dataset (A) and regular evaluation of model performance across demographic subgroups (B). Option C (deploying in multiple regions) affects latency or availability, not bias. Option D (larger neural network) does not address data imbalance. Option E (Amazon Rekognition's content moderation) is for detecting inappropriate content, not for bias mitigation.

Answer

Deploy the model in multiple regions

Answer

Use a larger neural network

Answer

Use Amazon Rekognition's content moderation

Question 9

An organization uses a third-party foundation model accessed through Amazon Bedrock. The compliance team requires that all model inputs and outputs be auditable and retained for one year. Which approach should the team implement?

Accepted Answer

Enable Bedrock model invocation logging and configure destination to S3 with a lifecycle policy to retain logs for one year. Bedrock model invocation logging captures all requests and responses and can be sent to CloudWatch Logs or S3 for retention. CloudTrail logs only API-level events (e.g., InvokeModel calls) but not the actual payloads. The other options do not provide payload logging.

Answer

Store all prompts and responses in a DynamoDB table via custom code in the application

Answer

Use VPC Flow Logs to capture network traffic to Bedrock

Answer

Enable CloudTrail data events for Bedrock

Question 10

A financial services company needs to ensure that the machine learning models used for loan approval are explainable and meet regulatory compliance. Which AWS feature can help explain model predictions?

Accepted Answer

SageMaker Clarify. SageMaker Clarify is the correct AWS service for explaining model predictions because it provides feature attribution and bias detection capabilities. It uses SHAP (SHapley Additive exPlanations) to generate explainability reports, which are essential for meeting regulatory compliance in financial services like loan approval.

Answer

SageMaker Ground Truth

Answer

SageMaker Automatic Model Tuning

Answer

SageMaker Model Monitor

Question 11

Which THREE are best practices for ensuring generated content complies with corporate brand guidelines when using Amazon Bedrock?

Accepted Answer

Implement guardrails to restrict tone, topics, and language. Option A is correct because Amazon Bedrock Guardrails let you define denied topics, word filters, and content filters that constrain the model's tone, subject matter, and language so outputs stay within corporate brand boundaries. Option B is correct because prompt engineering—embedding explicit instructions about brand voice, style, and formatting in the prompt—directly steers the model toward compliant outputs without retraining. Option E is correct because fine-tuning a model on a curated dataset of brand-compliant content adapts the model's behavior to consistently reproduce the organization's approved tone and terminology. Option C is not correct because raising the temperature increases randomness and creativity, which makes outputs less predictable and more likely to drift from brand guidelines. Option D is not correct because using random prompts to test variability does not enforce compliance; it merely measures output variation and does nothing to align content with brand standards.

Answer

Increase the temperature for more creative outputs

Answer

Use random prompts to test variability

Question 12

A company deploys a deep learning model for image classification using Amazon SageMaker. They are concerned about adversarial attacks that could misclassify images with small perturbations. Which of the following is the most effective approach to improve model robustness?

Accepted Answer

Apply adversarial training. Adversarial training is the most effective approach because it explicitly augments the training dataset with adversarial examples—inputs crafted with small, intentional perturbations designed to fool the model. By training on these perturbed samples, the model learns to recognize and resist such attacks, directly improving its robustness against adversarial perturbations in image classification tasks on SageMaker.

Answer

Reduce training data size

Answer

Use early stopping during training

Answer

Increase model complexity

Question 13

A data science team is fine-tuning a foundation model on Amazon SageMaker. Which THREE steps are part of the best practice? (Choose three.)

Accepted Answer

Monitor for catastrophic forgetting during fine-tuning.. Option B is correct because fine-tuning a foundation model on a narrow dataset can cause catastrophic forgetting, where the model loses previously learned general capabilities; monitoring for this (e.g., via evaluation on held-out general benchmarks) is a recognized best practice. Option C is correct because early stopping halts training when validation performance stops improving, which prevents overfitting to the fine-tuning dataset and is a standard SageMaker technique (e.g., via the EarlyStopping callback or stopping criterion). Option E is correct because using a diverse dataset that represents various scenarios improves generalization and reduces bias, ensuring the fine-tuned model performs well across the range of real-world inputs it will encounter. Option A is not correct because simply increasing model size does not guarantee better fine-tuning results and can increase cost, latency, and overfitting risk. Option D is not correct because deploying immediately after fine-tuning skips essential validation, evaluation, and testing steps before production release.

Answer

Increase model size to improve performance.

Answer

Deploy the model to production immediately after fine-tuning.

Question 14

A data science team is using Amazon SageMaker to build a model. They want to ensure that only authorized users can deploy models to production. Which TWO methods can they use to enforce this?

Accepted Answer

Use SageMaker Model Registry to require approval before deployment.. SageMaker Model Registry allows you to set up an approval workflow for model versions. By requiring explicit approval before a model can be deployed to production, you enforce a governance gate that prevents unauthorized or unverified models from being used in production endpoints.

Answer

Enable multi-factor authentication (MFA) for all AWS accounts.

Answer

Use AWS CloudTrail to audit deployment actions.

Answer

Use Amazon GuardDuty to monitor for unauthorized deployment.

Question 15

Which AWS service can be used to create human review workflows for high-risk AI predictions, ensuring a human-in-the-loop?

Accepted Answer

Amazon Augmented AI (A2I). Amazon Augmented AI (A2I) is the service designed to set up human review workflows for AI predictions, enabling human oversight when needed.

Answer

Amazon SageMaker Clarify

Answer

AWS Glue

Answer

AWS Step Functions