Courseiva

AWS Certified AI Practitioner AIF-C01 (AIF-C01) — Questions 76–150

619 questions total · 9pages · All types, answers revealed

Take a mock examExam hub
Page 1

Page 2 of 9

Page 3
76
Multi-Selectmedium

A company is deploying a large language model (LLM) for customer support. They want to reduce the risk of hallucinations. Which TWO approaches should they implement? (Choose two.)

Select 2 answers
A.Implement Amazon Bedrock Guardrails to define topics the model should avoid
B.Use a larger model without any retrieval mechanism
C.Fine-tune the model on a dataset of hallucinated examples
D.Increase the maximum token length to give the model more room to elaborate
E.Use Retrieval-Augmented Generation (RAG) to provide factual context
AnswersA, E

Guardrails can filter out responses on topics that are not supported by the knowledge base, reducing off-topic hallucinations.

Why this answer

RAG grounds the model in retrieved documents, reducing hallucinations. Bedrock Guardrails provide content filtering and topic restrictions. Fine-tuning on hallucinated data is counterproductive, and longer prompts can increase hallucinations.

Full explanation →
77
MCQmedium

A data scientist is training a binary classification model to predict customer churn. The dataset has 10,000 records with 9,500 non-churners and 500 churners. After training a logistic regression model, the model achieves 95% accuracy on the test set. However, the business team reports that the model is not useful because it predicts almost all customers as non-churners. Which metric should the data scientist use to evaluate the model's performance in this scenario?

A.Accuracy
B.R-squared
C.Precision
D.Recall
AnswerD

Recall measures the proportion of actual churners correctly identified, which is the key metric for this imbalanced problem.

Why this answer

(Recall) is correct because in this highly imbalanced dataset (95% non-churners vs 5% churners), the model's 95% accuracy is misleading—it can achieve this by simply predicting the majority class (non-churner) for all samples. Recall measures the proportion of actual churners correctly identified (True Positives / (True Positives + False Negatives)), directly addressing the business need to detect churn. A high recall ensures the model captures most churners, even at the cost of some false positives.

Exam trap

The AIF-C01 exam often tests the misconception that high accuracy always indicates a good model, especially in imbalanced datasets, leading candidates to overlook metrics like recall or precision that better reflect model utility for the specific business problem.

How to eliminate wrong answers

Option A is wrong because accuracy is a poor metric for imbalanced datasets; a model that predicts all samples as the majority class can achieve high accuracy (95% here) while failing to identify any churners, making it useless for the business goal. Option B is wrong because R-squared is a metric for regression models, measuring the proportion of variance explained by the independent variables, and is not applicable to binary classification tasks like churn prediction. Option C is wrong because precision (True Positives / (True Positives + False Positives)) focuses on the correctness of positive predictions; while important, it does not capture the model's ability to find all churners—a model with high precision but low recall might still miss most churners, which is the core issue reported by the business team.

Full explanation →
78
MCQmedium

A developer is using Amazon Bedrock to generate product descriptions. The developer notices that the model sometimes outputs descriptions that contradict the provided product specifications. Which parameter adjustment would MOST directly reduce factual inconsistencies?

A.Increase maxTokens to allow longer descriptions
B.Decrease temperature to a value close to 0
C.Increase topP to 1.0
D.Set topK to a higher value
AnswerB

Lower temperature makes the model more deterministic and less likely to deviate from the given specifications.

Why this answer

Decreasing temperature to a value close to 0 makes the model more deterministic and less creative, which reduces the likelihood of generating random or contradictory content. In Amazon Bedrock, temperature controls the randomness of token selection; lower values cause the model to choose the most probable tokens, aligning outputs more closely with the provided product specifications and minimizing factual inconsistencies.

Exam trap

AWS often tests the misconception that increasing output length (maxTokens) or expanding token selection (topP, topK) improves accuracy, when in fact these parameters increase variability and the risk of factual errors, whereas lowering temperature is the direct control for reducing randomness.

How to eliminate wrong answers

Option A is wrong because increasing maxTokens only extends the maximum length of the generated text, which does not address the randomness or creativity that leads to contradictions; it may even allow more room for errors. Option C is wrong because increasing topP to 1.0 includes all possible tokens up to the cumulative probability threshold, which actually increases diversity and can worsen factual inconsistencies by allowing less probable tokens. Option D is wrong because setting topK to a higher value expands the pool of candidate tokens considered, increasing randomness and the chance of generating contradictory content, rather than reducing it.

Full explanation →
79
MCQhard

A healthcare company needs to use Amazon SageMaker Ground Truth for data labeling. The data includes protected health information (PHI) that must remain in the US. Which configuration meets the compliance requirements?

A.Use a vendor-managed workforce and set up data encryption
B.Use a private workforce consisting of the company's employees and launch the labeling job in the us-east-1 region
C.Use a public workforce (Mechanical Turk) and select the US East region
D.Use a private workforce but launch the labeling job in the eu-west-1 region
AnswerB

Private workforce ensures data is handled by employees under the company's control, and region choice ensures data residency.

Why this answer

A private workforce consisting of the company's own employees ensures that PHI is never exposed to external workers, and launching the labeling job in us-east-1 keeps all data within the US, satisfying the data residency requirement. Amazon SageMaker Ground Truth allows you to restrict data access to a private workforce that you manage, and by selecting a US region, you ensure that data processing and storage remain within US borders.

Exam trap

The trap here is that candidates often assume that selecting a US region with a public workforce is sufficient for compliance, overlooking that PHI must not be exposed to external workers regardless of geographic location.

How to eliminate wrong answers

Option A is wrong because a vendor-managed workforce involves third-party vendors who may not have the same compliance controls for PHI, and data encryption alone does not guarantee that data remains within the US. Option C is wrong because a public workforce (Mechanical Turk) exposes PHI to anonymous external workers, which violates HIPAA and data privacy requirements, even if the region is set to US East. Option D is wrong because launching the labeling job in eu-west-1 (Ireland) violates the requirement that PHI must remain in the US, as data would be processed and stored in the European Union.

Full explanation →
80
MCQmedium

A media company is using Amazon Bedrock to generate captions for images. They have a batch processing pipeline that sends thousands of images daily to the Bedrock API using the Titan Image Generator G1 model. Recently, they started receiving ThrottlingException errors during peak hours. The team needs to process all images within 24 hours without changing the model or the application code. The current account has a default quota of 10 requests per second (RPS) for the Titan model in us-east-1. The team estimates they need 50 RPS during peak hours. They have already implemented exponential backoff in the client, but the errors persist. What is the MOST effective solution to resolve the throttling issue?

A.Request a service quota increase for the InvokeModel API for the Titan model in us-east-1
B.Use Amazon SageMaker batch transform to process images offline
C.Distribute the requests across multiple AWS Regions
D.Switch to a different foundation model that has a higher default quota
AnswerA

Increasing quota directly resolves throttling.

Why this answer

The team has already implemented exponential backoff, but the errors persist because their current quota of 10 RPS is insufficient for the required 50 RPS. Requesting a service quota increase for the InvokeModel API for the Titan Image Generator G1 model in us-east-1 directly addresses the root cause by raising the throughput limit, allowing the existing application code and model to handle the peak load without any architectural changes.

Exam trap

The trap here is that candidates may think exponential backoff or distributing across Regions solves all throttling, but the core issue is a hard service quota that must be increased, not a transient rate limit.

How to eliminate wrong answers

Option B is wrong because Amazon SageMaker batch transform is designed for offline inference on SageMaker endpoints, not for invoking Bedrock APIs; it would require changing the application code and infrastructure, which the question explicitly prohibits. Option C is wrong because distributing requests across multiple AWS Regions would require modifying the application code to route traffic to different endpoints, and it does not address the underlying quota issue in the primary region; it also introduces latency and complexity. Option D is wrong because switching to a different foundation model would require changing the application code and potentially the image generation logic, which is not allowed; moreover, other models may have different default quotas or capabilities, and the goal is to process images with the Titan model.

Full explanation →
81
MCQmedium

A startup is using Amazon Bedrock to power a virtual assistant. They need to ensure that personally identifiable information (PII) is not included in the model's responses. Which feature should they enable?

A.Enable PII redaction in the Bedrock guardrails.
B.Enable model invocation logging.
C.Configure a VPC endpoint.
D.Enable data encryption at rest.
AnswerA

Guardrails can redact PII from prompts and completions.

Why this answer

Amazon Bedrock Guardrails provide a configurable content filtering and PII redaction feature that can automatically detect and mask personally identifiable information (PII) in model inputs and outputs. By enabling PII redaction within guardrails, the startup can ensure that sensitive data like names, addresses, or credit card numbers are removed or obfuscated before the virtual assistant's responses reach the user. This is the direct and intended mechanism for preventing PII leakage in model responses.

Exam trap

The trap here is that candidates often confuse data protection features like encryption or logging with content filtering, not realizing that PII redaction is a specific guardrail policy that actively modifies model outputs in real time.

How to eliminate wrong answers

Option B is wrong because model invocation logging captures metadata and request/response payloads for auditing and debugging, but it does not actively redact or filter PII from responses — it only records what was sent and received. Option C is wrong because configuring a VPC endpoint provides private network connectivity to Bedrock without traversing the public internet, but it has no capability to inspect or modify the content of model responses for PII. Option D is wrong because enabling data encryption at rest protects stored data (e.g., logs, model artifacts) from unauthorized access, but it does not perform real-time redaction of PII in model outputs during inference.

Full explanation →
82
MCQeasy

A developer is building an application that generates product descriptions from images using a multimodal model. Which AWS service provides access to multimodal foundation models?

A.Amazon Rekognition
B.Amazon Textract
C.Amazon Comprehend
D.Amazon Bedrock
AnswerD

Bedrock provides access to foundation models, including multimodal models that can generate text from images.

Why this answer

Amazon Bedrock is a managed service that provides access to a wide range of foundation models (FMs) from leading AI providers, including multimodal models that can process both images and text to generate product descriptions. This makes it the correct choice for building an application that requires multimodal capabilities.

Exam trap

The trap here is that candidates may confuse AWS AI services that handle specific modalities (Rekognition for images, Comprehend for text) with Bedrock, which is the only service that provides access to generative multimodal foundation models capable of combining both modalities in a single inference.

How to eliminate wrong answers

Option A is wrong because Amazon Rekognition is a computer vision service for image and video analysis (e.g., object detection, facial recognition), but it does not provide access to generative multimodal foundation models. Option B is wrong because Amazon Textract is an OCR service that extracts text from documents, not a platform for accessing or running multimodal generative models. Option C is wrong because Amazon Comprehend is a natural language processing (NLP) service for text analysis (e.g., sentiment, entities), and it lacks support for multimodal input or generative model access.

Full explanation →
83
MCQmedium

A machine learning engineer wants to ensure that a SageMaker notebook instance only has access to a specific S3 bucket containing training data. The notebook instance is in a VPC. What is the most secure way to restrict access?

A.Use a VPC endpoint for S3 and a bucket policy that restricts access to the VPC endpoint.
B.Place the notebook instance in a private subnet with a NAT gateway.
C.Use AWS KMS to encrypt the bucket and grant the notebook role decrypt permissions.
D.Assign an IAM role to the notebook with an S3 bucket policy that only allows access to that bucket.
AnswerA

Combines network-level and resource-based policy to enforce access only from the VPC.

Why this answer

Using a VPC endpoint for S3 combined with a bucket policy that restricts access to that specific endpoint ensures that only traffic originating from within the VPC (and thus from the SageMaker notebook instance) can reach the S3 bucket. This approach enforces network-level isolation and prevents access from any other source, including the public internet, even if the IAM role is compromised. It is the most secure method because it layers network policy (VPC endpoint) with resource-based policy (bucket policy) to create a tightly scoped access control.

Exam trap

The trap here is that candidates often think IAM roles and bucket policies alone are sufficient for security, but they overlook the need for network-level restrictions (like VPC endpoints) to prevent data exfiltration or unauthorized access from outside the VPC.

How to eliminate wrong answers

Option B is wrong because placing the notebook instance in a private subnet with a NAT gateway still allows outbound internet traffic through the NAT gateway, which does not restrict access to a specific S3 bucket; the notebook could potentially reach any S3 bucket or internet resource, and the NAT gateway does not enforce bucket-level restrictions. Option C is wrong because using AWS KMS to encrypt the bucket and granting the notebook role decrypt permissions only protects data at rest but does not restrict which S3 bucket the notebook can access; the notebook could still access any bucket if the IAM policy allows it. Option D is wrong because assigning an IAM role with a bucket policy that only allows access to that bucket is a necessary but insufficient security measure; it does not prevent the notebook from accessing the bucket from outside the VPC or via the public internet, and it lacks the network-level restriction provided by a VPC endpoint.

Full explanation →
84
MCQhard

A financial services company needs to use a foundation model for sensitive data analysis. They require that all data remains within a VPC and no data leaves the AWS network. Which solution should they choose?

A.Use Amazon Comprehend with VPC.
B.Use Amazon Bedrock with a public endpoint.
C.Use Amazon Bedrock with a custom model and VPC endpoints.
D.Use Amazon SageMaker with a VPC-only real-time endpoint hosting a foundation model.
AnswerC

Bedrock custom models support VPC endpoints to keep data within the network.

Why this answer

Amazon Bedrock with a custom model and VPC endpoints ensures that all data remains within the VPC and never traverses the public internet, meeting the requirement for sensitive data analysis. VPC endpoints (AWS PrivateLink) allow private connectivity to Bedrock, and a custom model can be deployed within the VPC for inference, keeping data entirely within the AWS network.

Exam trap

The trap here is that candidates may confuse SageMaker's VPC hosting capabilities with Bedrock's managed service model, or assume that any AWS service with VPC support (like Comprehend) can serve as a foundation model solution, when in fact Bedrock is the only managed service designed for foundation model access with private VPC endpoints.

How to eliminate wrong answers

Option A is wrong because Amazon Comprehend is a natural language processing service that does not provide foundation model capabilities for generative AI tasks, and its VPC support only applies to data processing, not to hosting or invoking foundation models. Option B is wrong because using Amazon Bedrock with a public endpoint means data and requests travel over the public internet, violating the requirement that no data leaves the AWS network. Option D is wrong because Amazon SageMaker with a VPC-only real-time endpoint hosting a foundation model would require you to self-manage the model and infrastructure, which is not the recommended approach for using a managed foundation model service like Bedrock, and it does not leverage Bedrock's VPC endpoint integration for private access.

Full explanation →
85
MCQmedium

A company deployed a chatbot using Amazon Lex integrated with a Lambda function that invokes Claude on Amazon Bedrock. The Lambda function retrieves relevant documents from an Amazon Kendra index to use as context. Users report that the chatbot's responses are often irrelevant or incorrect despite the Kendra index containing accurate information. The logs show that the Lambda function is correctly passing retrieved documents to the model. What is the most likely cause and solution?

A.Switch to a larger foundation model like Claude 3 Opus
B.The model's temperature is set too high; reduce it to 0.1
C.The maximum tokens limit is too low; increase it to 4096
D.The chunking strategy for documents is too coarse or inappropriate; refine chunking and use semantic search in Kendra
AnswerD

Proper chunking ensures each chunk contains coherent information relevant to potential queries; Kendra's semantic search improves relevance.

Why this answer

The issue likely stems from the chunking and retrieval strategy. If the retrieved document chunks do not contain the exact answer or are poorly segmented, the model may not have the necessary context. Improving chunking to be more semantic and ensuring retrieval uses a relevant similarity metric (e.g., using Kendra's relevance tuning) would help.

Increasing temperature or reducing tokens would degrade quality. Switching model may not address the root cause.

Full explanation →
86
MCQhard

A security team is concerned about adversarial attacks on their image classification model deployed on Amazon SageMaker. They want to test robustness against carefully crafted inputs that cause misclassification. What approach should they use?

A.Data augmentation on the training set
B.A/B testing between two similar models
C.SageMaker Model Monitor with adversarial drift
D.Generating adversarial examples using SageMaker Clarify
AnswerD

Clarify includes adversarial validation capabilities to test robustness.

Why this answer

SageMaker Clarify includes built-in capabilities for generating adversarial examples to evaluate model robustness against inputs designed to cause misclassification. This directly addresses the security team's concern by simulating adversarial attacks, allowing them to measure and improve the model's resilience before deployment.

Exam trap

AWS often tests the distinction between monitoring for natural data drift (Model Monitor) and proactively testing for adversarial robustness (Clarify), leading candidates to mistakenly choose Model Monitor when the question explicitly asks for testing against crafted inputs.

How to eliminate wrong answers

Option A is wrong because data augmentation improves generalization to natural variations, not adversarial robustness against crafted perturbations that exploit model vulnerabilities. Option B is wrong because A/B testing compares model performance on normal traffic, not adversarial inputs, and does not generate or test against adversarial examples. Option C is wrong because SageMaker Model Monitor detects data drift and quality issues in production, not adversarial attacks; 'adversarial drift' is not a standard feature and does not involve generating adversarial examples.

Full explanation →
87
Multi-Selectmedium

A company is building a generative AI application using Amazon Bedrock and needs to ensure that the model does not generate outputs containing personally identifiable information (PII). Which TWO actions should the company take? (Choose 2)

Select 2 answers
A.Implement a custom AWS Lambda function to scan and redact PII from inputs and outputs.
B.Use AWS Identity and Access Management (IAM) policies to restrict model access.
C.Enable Amazon CloudWatch Logs to capture and audit model outputs.
D.Configure Amazon Bedrock Guardrails to block or mask PII.
E.Place the Bedrock model endpoint within a private VPC.
AnswersA, D

Lambda can use PII detection libraries to filter sensitive data.

Why this answer

A custom AWS Lambda function can be integrated into the application workflow to programmatically scan and redact PII from both inputs and outputs before they reach or leave the Bedrock model. This provides a flexible, code-driven approach to data sanitization, allowing the use of libraries like Amazon Comprehend or regex patterns to detect and mask PII entities such as names, addresses, and social security numbers.

Exam trap

The AIF-C01 exam often tests the distinction between network-level security controls (like VPCs) and content-level data protection mechanisms, leading candidates to mistakenly choose VPC isolation as a solution for PII redaction.

Full explanation →
88
MCQhard

A company wants to use a large language model to generate code based on natural language descriptions. They need to minimize latency and control costs by running inference on their own infrastructure. Which approach is most suitable?

A.Use Amazon Bedrock API
B.Use Amazon SageMaker to deploy a custom LLM
C.Use Amazon Comprehend
D.Use Amazon Lex
AnswerB

SageMaker can deploy models on customer-specified instances, giving control over latency and cost.

Why this answer

Amazon SageMaker allows you to deploy a custom large language model (LLM) on your own infrastructure, giving you full control over inference latency and cost. By using SageMaker endpoints with auto-scaling and instance selection, you can optimize for low-latency responses while avoiding per-token API charges from managed services.

Exam trap

AWS often tests the distinction between managed API services (like Bedrock) and self-managed deployment options (like SageMaker), where candidates mistakenly choose Bedrock for 'control' over costs and latency, not realizing that Bedrock is a pay-per-token managed service with no infrastructure control.

How to eliminate wrong answers

Option A is wrong because Amazon Bedrock is a managed API service that charges per-token and does not allow you to run inference on your own infrastructure, so you cannot control latency or costs at the infrastructure level. Option C is wrong because Amazon Comprehend is a natural language processing (NLP) service for tasks like sentiment analysis and entity extraction, not a generative AI service capable of code generation from natural language. Option D is wrong because Amazon Lex is designed for building conversational chatbots using intent-based models, not for deploying large language models for code generation.

Full explanation →
89
MCQmedium

Refer to the exhibit. A data scientist runs an Amazon SageMaker Clarify bias analysis on a binary classifier. The pre-training ClassImbalance is 1.5 and the post-training DPPL is 0.15. What should the data scientist conclude?

A.The data is highly imbalanced and the model is unbiased.
B.The data has a mild class imbalance, but the model shows a noticeable bias in predictions.
C.The pre-training metric indicates a fairness issue, but the post-training metric is acceptable.
D.The data is perfectly balanced and the model is fair.
AnswerB

ClassImbalance of 1.5 is moderate; DPPL of 0.15 indicates a 15% difference, which is concerning.

Why this answer

The pre-training ClassImbalance metric of 1.5 indicates a mild class imbalance (values close to 1.0 indicate balance, while values significantly above 1.0 indicate imbalance). The post-training DPPL (Difference in Positive Proportions in Labels) metric of 0.15 exceeds the commonly accepted fairness threshold of 0.10, indicating a noticeable bias in the model's predictions. Therefore, the data has a mild imbalance, but the model exhibits a bias that warrants further investigation.

Exam trap

In AWS AI Practitioner exams, a common misconception is that a low pre-training imbalance automatically means the model is fair, but the post-training DPPL metric directly measures prediction bias and can reveal unfairness even when the data appears balanced.

How to eliminate wrong answers

Option A is wrong because a ClassImbalance of 1.5 indicates a mild imbalance, not a highly imbalanced dataset, and the DPPL of 0.15 suggests the model is biased, not unbiased. Option C is wrong because the pre-training metric of 1.5 does not indicate a fairness issue—it only measures class distribution, not fairness—and the post-training DPPL of 0.15 is above the 0.10 threshold, making it unacceptable. Option D is wrong because a ClassImbalance of 1.5 is not perfectly balanced (perfect balance is 1.0), and a DPPL of 0.15 indicates the model is not fair.

Full explanation →
90
MCQmedium

Refer to the exhibit. You receive this response from Amazon Bedrock. What is the most likely cause of the incomplete information?

A.The max_tokens limit was reached
B.The prompt was too short
C.The temperature was too high
D.The model lacks knowledge about capitals
AnswerA

stop_reason: max_tokens indicates the output was capped by the token limit.

Why this answer

The response from Amazon Bedrock shows an incomplete sentence that cuts off mid-thought, which is a classic symptom of hitting the max_tokens limit. When the generated output reaches the specified maximum number of tokens, the model stops generating immediately, resulting in truncated text. This is the most likely cause because the output is syntactically incomplete but otherwise coherent up to the cutoff point.

Exam trap

AWS often tests the distinction between output truncation (max_tokens) and output quality issues (temperature, prompt engineering), so the trap here is that candidates may incorrectly attribute a truncated response to model ignorance or randomness rather than the explicit token limit.

How to eliminate wrong answers

Option B is wrong because the prompt length does not directly cause incomplete output; a short prompt can still produce a complete response if the max_tokens limit is high enough. Option C is wrong because temperature controls randomness and creativity, not the length or truncation of the output; high temperature might produce less coherent text but would not cut off mid-sentence. Option D is wrong because the model's lack of knowledge about capitals would result in incorrect or hallucinated information, not a truncated or incomplete sentence.

Full explanation →
91
MCQeasy

An organization wants to control which topics their AI chatbot can discuss. For example, they want to block all conversations about investment advice. Which Amazon Bedrock Guardrails feature should they configure?

A.Contextual grounding check
B.Content filtering with category-based harmful content filters
C.Sensitive information filters
D.Topic restrictions
AnswerD

Topic restrictions explicitly define which topics are allowed or denied.

Why this answer

Topic restrictions allow you to define denied topics. When the model's input or output matches a denied topic, the guardrail blocks the interaction.

Full explanation →
92
MCQeasy

A company is using Amazon SageMaker to train machine learning models on sensitive customer data. Which AWS service can be used to encrypt the data at rest in the S3 bucket used by SageMaker?

A.AWS Key Management Service (KMS)
B.AWS CloudHSM
C.AWS Secrets Manager
D.AWS Certificate Manager (ACM)
AnswerA

KMS creates and manages encryption keys used for S3 server-side encryption.

Why this answer

AWS Key Management Service (KMS) is the correct service because it provides managed encryption keys that can be used to enable server-side encryption (SSE-KMS) for Amazon S3 buckets. When SageMaker accesses training data from S3, it can use a customer-managed KMS key to encrypt data at rest, ensuring sensitive customer data remains protected. KMS integrates directly with S3 and SageMaker, allowing you to specify a KMS key in the SageMaker training job configuration.

Exam trap

The trap here is that candidates often confuse AWS CloudHSM with KMS, thinking that a dedicated HSM is required for encryption, but KMS is the simpler, fully managed service that directly integrates with S3 and SageMaker for at-rest encryption.

How to eliminate wrong answers

Option B (AWS CloudHSM) is wrong because CloudHSM provides dedicated hardware security modules for key generation and storage, but it does not directly integrate with S3 for server-side encryption; you would need to manage the encryption process yourself, making it more complex and less suitable for simple at-rest encryption. Option C (AWS Secrets Manager) is wrong because Secrets Manager is designed to securely store and rotate secrets like database credentials and API keys, not to manage encryption keys for S3 data at rest. Option D (AWS Certificate Manager (ACM)) is wrong because ACM is used to provision, manage, and deploy SSL/TLS certificates for securing network traffic (in transit), not for encrypting data at rest in S3.

Full explanation →
93
Multi-Selectmedium

A hospital is deploying an AI system to assist in diagnosing diseases from medical images. According to the EU AI Act, this system may be classified as high-risk. Which THREE requirements should the hospital address to comply with the EU AI Act for high-risk AI systems?

Select 3 answers
A.Ensure transparency and provision of information to users
B.Establish a risk management system throughout the lifecycle
C.Enable human oversight to prevent or minimize risks
D.Use only open-source models
E.Deploy the system without any testing in a sandbox environment
AnswersA, B, C

High-risk systems must be transparent about capabilities and limitations.

Why this answer

The EU AI Act requires high-risk systems to have risk management, transparency, and human oversight among other requirements.

Full explanation →
94
MCQeasy

A social media platform uses an AI system to moderate content. They want to ensure that human reviewers can review decisions when the AI is uncertain. Which AWS service can be used to set up a human review workflow for AI predictions?

A.Amazon Mechanical Turk
B.AWS Lambda
C.Amazon SageMaker Ground Truth
D.Amazon Augmented AI (A2I)
AnswerD

A2I provides a managed service to create human review loops, integrating with SageMaker and other AWS services to route predictions to humans for review.

Why this answer

Amazon Augmented AI (A2I) enables easy integration of human review workflows into ML applications, allowing humans to review low-confidence predictions or edge cases.

Full explanation →
95
MCQeasy

Refer to the exhibit. A developer wants to ensure the notebook instance can access the internet to download packages. Which property configuration ensures this?

A.DirectInternetAccess: Enabled
B.VolumeSizeInGB: 5
C.InstanceType: ml.t2.medium
D.The resource type AWS::SageMaker::NotebookInstance
AnswerA

Setting DirectInternetAccess to Enabled allows the notebook instance to access the internet.

Why this answer

Setting `DirectInternetAccess: Enabled` on an AWS SageMaker notebook instance allows it to access the internet through a VPC with a Network Address Translation (NAT) gateway or via the public internet if the instance is not in a VPC. This configuration is required to download packages from external repositories like PyPI or conda.

Exam trap

AWS often tests the distinction between resource type identifiers and configurable properties, so candidates may mistakenly think that specifying `AWS::SageMaker::NotebookInstance` as the resource type itself enables internet access, rather than recognizing it as a CloudFormation resource declaration.

How to eliminate wrong answers

Option B is wrong because `VolumeSizeInGB: 5` only specifies the size of the Amazon EBS storage volume attached to the notebook instance, which does not affect internet connectivity. Option C is wrong because `InstanceType: ml.t2.medium` defines the compute capacity (CPU and memory) of the instance, not its network access capabilities. Option D is wrong because `AWS::SageMaker::NotebookInstance` is the resource type identifier in AWS CloudFormation, not a property that controls internet access.

Full explanation →
96
Multi-Selecteasy

Which TWO of the following are benefits of using Amazon Bedrock for building applications with foundation models?

Select 2 answers
A.No infrastructure management
B.Automatic model fine-tuning
C.Access to multiple foundation models
D.Free tier for all models
E.Built-in image generation capability
AnswersA, C

Bedrock is serverless; AWS handles the underlying infrastructure.

Why this answer

Amazon Bedrock is a fully managed service that abstracts away the underlying infrastructure required to host and run foundation models (FMs). By using Bedrock, you do not need to provision, configure, or manage servers, GPUs, or scaling policies, which is a key benefit for developers who want to focus on building applications rather than managing infrastructure. Additionally, Bedrock provides a single API to access multiple FMs from providers like AI21 Labs, Anthropic, Cohere, Meta, and Stability AI, enabling you to choose the best model for your use case without managing separate endpoints or integrations.

Exam trap

AWS often tests the misconception that Amazon Bedrock includes built-in capabilities like automatic fine-tuning or image generation, when in reality these are model-specific features that you must explicitly select and configure, not inherent service features.

Full explanation →
97
MCQmedium

A data scientist is training a model using Amazon SageMaker and notices the training loss is decreasing but validation loss starts increasing after a few epochs. Which technique should they apply to address this?

A.Increase batch size
B.Increase the learning rate
C.Add more training data
D.Add regularization (e.g., L1 or L2)
AnswerD

Regularization penalizes large weights and reduces overfitting, which is indicated by increasing validation loss.

Why this answer

The scenario describes overfitting, where the model memorizes training data but fails to generalize to validation data. Adding regularization (L1 or L2) penalizes large weights, reducing model complexity and improving generalization. This is a standard technique in SageMaker training jobs, often configured via the `regularizer` hyperparameter in frameworks like TensorFlow or MXNet.

Exam trap

The trap here is that candidates confuse overfitting with underfitting or optimization issues, and incorrectly choose to increase learning rate or batch size, not recognizing that rising validation loss with falling training loss is the classic signature of overfitting.

How to eliminate wrong answers

Option A is wrong because increasing batch size typically stabilizes gradient estimates but does not directly address overfitting; it may even reduce generalization by sharpening minima. Option B is wrong because increasing the learning rate can cause divergence or overshooting of the loss minimum, worsening both training and validation loss. Option C is wrong because adding more training data can help generalization but is not a direct fix for overfitting when validation loss increases; it may not be feasible or sufficient, and regularization is the immediate corrective action.

Full explanation →
98
MCQhard

A company is using Amazon Bedrock to generate product descriptions. They notice that the model sometimes produces descriptions that contain factual errors about the products. Which TWO actions should they take to improve factual accuracy?

A.Implement Retrieval Augmented Generation (RAG) with a product knowledge base
B.Reduce the temperature parameter to 0.1
C.Use a curated prompt with few-shot examples of accurate descriptions
D.Increase the max_tokens to allow longer descriptions
E.Use human reviewers to correct errors after generation
AnswerA, C

RAG provides current, accurate information to the model.

Why this answer

Retrieval Augmented Generation (RAG) grounds the model's output in a curated product knowledge base, allowing it to retrieve and cite authoritative facts during generation. This directly reduces hallucinations by ensuring the model references verified data rather than relying solely on its parametric memory.

Exam trap

The AIF-C01 exam often tests the misconception that tuning generation parameters (like temperature or max_tokens) can fix factual accuracy, when in reality only grounding techniques like RAG or curated few-shot examples address the underlying hallucination problem.

How to eliminate wrong answers

Option B is wrong because reducing the temperature parameter to 0.1 makes the model more deterministic and repetitive, but it does not introduce factual grounding—it only reduces randomness, which can still produce plausible-sounding but incorrect facts. Option D is wrong because increasing max_tokens allows longer descriptions but does not improve factual accuracy; it may even increase the chance of generating more hallucinated content. Option E is wrong because human reviewers after generation is a validation step, not a method to improve the model's factual accuracy at inference time; it adds latency and cost without addressing the root cause of factual errors.

Full explanation →
99
Multi-Selectmedium

A company is deploying a generative AI model on Amazon Bedrock and needs to monitor for potential misuse. Which THREE measures should they implement? (Choose 3)

Select 3 answers
A.Require multi-factor authentication (MFA) for all API calls.
B.Configure Amazon Bedrock Guardrails to block harmful content.
C.Use AWS CloudTrail to log API calls and Amazon Bedrock actions.
D.Place the Bedrock endpoint in a private VPC with no internet access.
E.Enable model invocation logging in Amazon CloudWatch.
AnswersB, C, E

Guardrails proactively filter inputs and outputs.

Why this answer

Amazon Bedrock Guardrails provides configurable content filters that can block harmful or undesirable content in both input prompts and model responses. This is a direct monitoring and prevention mechanism for misuse, allowing administrators to define policies for topics, toxicity, and sensitive information.

Exam trap

The AIF-C01 exam often tests the distinction between security controls that prevent access (like MFA or VPC isolation) versus monitoring controls that detect or block misuse at the content level, leading candidates to confuse network security with content safety.

Full explanation →
100
Multi-Selectmedium

A company uses Amazon Bedrock and wants to ensure that the model outputs are grounded in a set of provided documents to reduce hallucinations. Which TWO actions should they take? (Select TWO.)

Select 2 answers
A.Enable the grounding check in Bedrock Guardrails
B.Configure a word filter to block ungrounded phrases
C.Enable model invocation logging to S3
D.Use Amazon Bedrock Knowledge Bases to store and retrieve document chunks
E.Fine-tune the model on the documents
AnswersA, D

Grounding check ensures responses are anchored to source documents.

Why this answer

Bedrock Guardrails' grounding check verifies that model responses are supported by source documents. Knowledge Bases for Amazon Bedrock provide a RAG architecture to retrieve relevant document chunks.

Full explanation →
101
MCQeasy

Which prompt engineering technique involves providing the model with a few examples of desired input-output pairs before asking it to complete a new instance?

A.Zero-shot prompting
B.Chain-of-thought prompting
C.System prompting
D.Few-shot prompting
AnswerD

Few-shot includes a small number of examples in the prompt.

Why this answer

Few-shot prompting (Option D) is the correct technique because it explicitly involves providing the model with a few examples of desired input-output pairs before asking it to complete a new instance. This helps the model understand the pattern, format, or task from the examples, improving performance on tasks like classification, translation, or formatting without requiring fine-tuning.

Exam trap

AWS often tests the distinction between zero-shot and few-shot prompting, where candidates may confuse 'providing examples' with 'setting system instructions' or 'step-by-step reasoning', leading them to incorrectly select chain-of-thought or system prompting.

How to eliminate wrong answers

Option A is wrong because zero-shot prompting asks the model to perform a task without any examples, relying solely on its pre-trained knowledge. Option B is wrong because chain-of-thought prompting involves guiding the model to reason step-by-step, often with intermediate reasoning steps, not providing input-output pairs. Option C is wrong because system prompting sets the overall behavior, persona, or constraints for the model (e.g., 'You are a helpful assistant'), but does not provide specific input-output examples for a task.

Full explanation →
102
MCQmedium

A developer is building a RAG-based Q&A bot with Amazon Bedrock Knowledge Bases. They need a managed vector store for document embeddings. Which service should they use?

A.Amazon OpenSearch Serverless
B.Amazon DynamoDB
C.Amazon RDS
D.Amazon S3
AnswerA

OpenSearch Serverless with k-NN plugin provides managed vector storage.

Why this answer

Amazon Bedrock Knowledge Bases requires a vector store to store and query document embeddings for Retrieval-Augmented Generation (RAG). Amazon OpenSearch Serverless provides a managed, scalable vector engine that supports k-NN (k-nearest neighbor) search, making it the correct choice for this use case. It integrates natively with Bedrock Knowledge Bases to handle embedding storage and similarity search without manual infrastructure management.

Exam trap

The trap here is that candidates may confuse Amazon DynamoDB or Amazon RDS as viable options because they can store data, but they lack native vector search capabilities required for RAG, leading to an incorrect choice.

How to eliminate wrong answers

Option B (Amazon DynamoDB) is wrong because it is a key-value and document database that does not natively support vector similarity search or k-NN indexing, making it unsuitable as a vector store for RAG. Option C (Amazon RDS) is wrong because it is a relational database service that lacks built-in vector search capabilities; while extensions like pgvector for PostgreSQL exist, Amazon RDS is not a managed vector store and would require custom implementation. Option D (Amazon S3) is wrong because it is an object storage service that cannot perform vector similarity queries; it can store raw documents but not embeddings in a searchable vector index.

Full explanation →
103
MCQmedium

A company is building a chatbot using Amazon Bedrock and wants to ensure that the model generates responses consistent with its brand voice. Which technique should be used to provide the model with examples of desired responses without fine-tuning the model?

A.Fine-tune the model on a dataset of brand-compliant conversations.
B.Use prompt chaining to break down the conversation into multiple steps.
C.Implement a Retrieval Augmented Generation (RAG) system with brand documents.
D.Include few-shot examples in the system prompt to demonstrate the desired tone.
AnswerD

In-context learning via few-shot examples guides model behavior without retraining.

Why this answer

Few-shot prompting allows you to provide the model with examples of desired responses directly in the system prompt, guiding the model's tone and style without modifying its underlying weights. This technique is ideal for brand voice consistency when fine-tuning is not an option, as it leverages in-context learning to influence output behavior.

Exam trap

AWS often tests the distinction between in-context learning (few-shot prompting) and fine-tuning, trapping candidates who confuse RAG (which retrieves facts) with style guidance, or who think prompt chaining is for tone control rather than task decomposition.

How to eliminate wrong answers

Option A is wrong because fine-tuning requires modifying the model's weights, which contradicts the requirement of not fine-tuning the model. Option B is wrong because prompt chaining is a technique for decomposing complex tasks into sequential steps, not for providing examples of desired tone or style. Option C is wrong because Retrieval Augmented Generation (RAG) retrieves external knowledge from documents to ground responses in facts, but it does not inherently teach the model the specific tone or brand voice; it augments context, not style.

Full explanation →
104
MCQhard

An e-commerce company stores user interaction logs in Amazon S3. They want to use machine learning to segment users based on purchasing behavior. Which unsupervised learning algorithm is most appropriate?

A.Linear regression
B.Random forest
C.K-means clustering
D.Neural network
AnswerC

Unsupervised algorithm that groups data into clusters based on similarity.

Why this answer

K-means clustering is the most appropriate unsupervised learning algorithm for segmenting users based on purchasing behavior because it groups data points into clusters based on feature similarity without requiring labeled training data. The e-commerce scenario involves discovering natural groupings (segments) in user interaction logs, which is a classic clustering task, and K-means efficiently partitions users into K distinct segments by minimizing within-cluster variance.

Exam trap

The AIF-C01 exam often tests the distinction between supervised and unsupervised learning by presenting a clustering problem and including supervised algorithms as distractors, leading candidates to mistakenly pick a familiar algorithm like random forest or linear regression without recognizing the lack of labeled data.

How to eliminate wrong answers

Option A is wrong because linear regression is a supervised learning algorithm used for predicting continuous numeric values (e.g., sales amount) from labeled data, not for discovering unlabeled user segments. Option B is wrong because random forest is a supervised ensemble learning method used for classification or regression on labeled datasets, and it cannot perform unsupervised segmentation without target labels. Option D is wrong because neural networks are typically used in supervised or reinforcement learning contexts; while they can be adapted for unsupervised tasks (e.g., autoencoders), they are not the most straightforward or appropriate choice for simple user segmentation compared to K-means clustering.

Full explanation →
105
MCQhard

A financial services company is subject to strict regulatory requirements. They plan to use generative AI to summarize customer interaction logs. Which combination of AWS services and configurations best ensures compliance while maintaining accuracy?

A.Deploy an open-source model on Amazon Bedrock in a local on-premises server.
B.Use Amazon Bedrock with a foundation model and public internet access without encryption.
C.Use Amazon SageMaker to host a fine-tuned model with a public API key.
D.Use Amazon Bedrock with a private VPC endpoint, AWS KMS encryption, and content filtering.
AnswerD

This configuration meets regulatory requirements for data privacy and content safety.

Why this answer

It combines a private VPC endpoint to keep all traffic within the AWS network (avoiding public internet exposure), AWS KMS encryption for data at rest and in transit, and content filtering to block sensitive or non-compliant outputs. This architecture meets strict regulatory requirements for data privacy and security while using Amazon Bedrock's managed foundation models for accurate summarization.

Exam trap

A common misconception is that encryption alone ensures compliance. However, the trap here is that public internet access (even with HTTPS) violates strict regulatory requirements that mandate private network connectivity (via VPC endpoints) and data residency controls.

How to eliminate wrong answers

Option A is wrong because deploying an open-source model on a local on-premises server does not use Amazon Bedrock (which is a fully managed AWS service) and introduces operational overhead, potential compliance gaps, and lacks AWS-native encryption and auditing. Option B is wrong because using public internet access without encryption exposes customer interaction logs to interception and violates regulatory mandates for data in transit security (e.g., TLS). Option C is wrong because using a public API key with Amazon SageMaker exposes the model endpoint to unauthorized access and lacks the private networking and encryption controls required for compliance.

Full explanation →
106
MCQmedium

A company wants to build a customer service chatbot that answers questions about their internal policy documents. The documents are updated monthly, and the team cannot afford to retrain a model each time. Which approach is MOST appropriate?

A.Fine-tune a base LLM on the policy documents monthly
B.Use Retrieval-Augmented Generation (RAG) with the policy documents indexed in a vector store
C.Train a custom model from scratch on the policy documents each month
D.Use a larger foundation model with a longer context window and paste all documents into each prompt
AnswerB

RAG retrieves relevant document chunks at query time, ensuring the chatbot always answers from the latest uploaded documents without any model retraining.

Why this answer

RAG (Retrieval-Augmented Generation) allows the LLM to retrieve relevant document sections at inference time, so knowledge stays current without retraining. The other options either require expensive retraining for each update or lack document grounding.

Full explanation →
107
MCQhard

A team trains a model using Amazon SageMaker built-in XGBoost. After training, they want to evaluate feature importance. Which SageMaker feature allows them to view this?

A.SageMaker Debugger
B.SageMaker Experiments
C.SageMaker Autopilot
D.SageMaker Model Monitor
AnswerA

Debugger can capture internal model states like feature importance.

Why this answer

SageMaker Debugger provides built-in monitoring and visualization capabilities, including the ability to capture feature importance metrics (e.g., gain, cover, weight) from XGBoost training jobs. It automatically saves these metrics to Amazon S3 and allows you to view them through the SageMaker Studio Debugger dashboard or by querying the saved tensors, enabling direct evaluation of feature importance without additional custom code.

Exam trap

The trap here is that candidates often confuse SageMaker Experiments' tracking of training metrics (like accuracy or loss) with the ability to view model-specific internals like feature importance, which is a Debugger capability.

How to eliminate wrong answers

Option B (SageMaker Experiments) is wrong because it is designed for tracking and comparing training runs (e.g., hyperparameters, metrics, artifacts) but does not natively capture or expose feature importance values from the model. Option C (SageMaker Autopilot) is wrong because it automates the end-to-end ML pipeline (data preprocessing, model selection, hyperparameter tuning) and provides feature importance only as part of its generated candidate definition notebooks, not as a direct, real-time feature during a custom XGBoost training job. Option D (SageMaker Model Monitor) is wrong because it focuses on detecting data drift and model quality degradation in production deployments, not on extracting feature importance from a trained model.

Full explanation →
108
MCQhard

A healthcare company needs to use a foundation model for analyzing medical records while complying with HIPAA. They plan to use Amazon Bedrock. What should they do to meet HIPAA requirements?

A.Use a model that is HIPAA eligible in a region that supports BAA
B.Implement access logging for all API calls
C.Encrypt data at rest and in transit
D.All of the above
AnswerD

All three are required for HIPAA compliance with Bedrock.

Why this answer

HIPAA compliance in Amazon Bedrock requires a combination of controls: using a HIPAA-eligible model in a region where AWS offers a Business Associate Addendum (BAA), enabling access logging for auditability, and encrypting data at rest and in transit. None of the individual options alone satisfy all HIPAA requirements; only the full set of controls ensures compliance.

Exam trap

The trap here is that candidates often pick a single security control (like encryption or logging) thinking it alone ensures HIPAA compliance, but the exam tests that HIPAA requires a combination of administrative, physical, and technical safeguards, all of which must be addressed.

How to eliminate wrong answers

Option A is wrong because while using a HIPAA-eligible model in a BAA-supported region is necessary, it does not address audit logging or encryption requirements. Option B is wrong because access logging alone provides audit trails but does not ensure the model is HIPAA-eligible or that data encryption is enforced. Option C is wrong because encrypting data at rest and in transit is critical but does not cover the need for a BAA or access logging.

All three are required together.

Full explanation →
109
MCQeasy

A company wants to evaluate the quality of a text generation model for a summarization task. They have reference summaries written by humans. Which automated metric compares the generated summary to the reference by measuring n-gram overlap?

A.BLEU
B.Perplexity
C.ROUGE
D.BERTScore
AnswerC

ROUGE measures recall of n-grams and is commonly used for summarization evaluation.

Why this answer

ROUGE (Recall-Oriented Understudy for Gisting Evaluation) measures n-gram overlap between generated and reference summaries. BLEU is for translation, BERTScore uses embeddings, and perplexity measures language model confidence.

Full explanation →
110
MCQmedium

A retail company wants to generate product descriptions from catalog data. The data includes structured attributes (e.g., price, brand) and unstructured reviews. The team needs to ensure factual accuracy. Which approach is most appropriate?

A.Use prompt engineering with few-shot examples
B.Fine-tune a foundation model on the entire product catalog
C.Deploy a larger foundation model with more parameters
D.Implement Retrieval-Augmented Generation (RAG) with a knowledge base
AnswerD

RAG retrieves relevant product data at inference time, ensuring factual accuracy and allowing updates without retraining.

Why this answer

Retrieval-Augmented Generation (RAG) retrieves relevant documents (product attributes, reviews) and provides them as context to the model, reducing hallucinations and grounding responses in facts.

Full explanation →
111
Multi-Selectmedium

A company wants to build an ML model to predict customer lifetime value. The dataset includes numerical features (age, income) and categorical features (gender, region). Which TWO preprocessing steps should be applied to the categorical features before training a linear regression model? (Choose TWO.)

Select 2 answers
A.Feature scaling
B.Imputation of missing values
C.Principal component analysis (PCA)
D.One-hot encoding
E.Label encoding
AnswersB, D

Imputation of missing values is a standard preprocessing step for categorical features to handle missing data, which linear regression cannot tolerate.

Why this answer

Imputation of missing values is necessary for categorical features because linear regression cannot handle missing data. Common practice is to impute with the mode. One-hot encoding converts categorical variables into binary vectors, avoiding ordinal bias that label encoding would introduce.

These two steps ensure the categorical data is suitable for linear regression.

Exam trap

AWS often tests the distinction between label encoding and one-hot encoding, trapping candidates who think label encoding is acceptable for linear models when it actually introduces ordinal bias.

Full explanation →
112
MCQmedium

A data scientist is using Amazon SageMaker Studio. The company policy requires that all data used in SageMaker Studio notebooks be encrypted at rest and in transit. Which configuration should be enabled to meet this requirement?

A.Store all data in encrypted DynamoDB tables and query from notebooks.
B.Use a VPC with no internet access and enable S3 encryption for all data.
C.Enable SageMaker Studio's default encryption using the AWS managed key for EFS.
D.Enable SageMaker Studio's data encryption using a customer managed key (CMK) and configure the notebook to use HTTPS for all traffic.
AnswerD

Ensures encryption at rest (CMK) and in transit (HTTPS).

Why this answer

It addresses both encryption at rest and in transit. SageMaker Studio uses Amazon EFS for persistent storage, and enabling encryption at rest with a customer managed key (CMK) ensures data on the EFS file system is encrypted. Additionally, configuring the notebook to use HTTPS ensures all traffic between the notebook and other services is encrypted in transit, meeting the company policy.

Exam trap

The trap here is that candidates often assume default encryption (Option C) is sufficient, but it only covers data at rest on EFS and ignores the encryption in transit requirement, which is explicitly tested in this scenario.

How to eliminate wrong answers

Option A is wrong because DynamoDB encryption only protects data at rest in DynamoDB, but SageMaker Studio notebooks do not natively use DynamoDB for storage; the primary storage is EFS, and this option does not address encryption in transit or the actual data storage used by Studio. Option B is wrong because while using a VPC with no internet access and enabling S3 encryption can protect data in S3, SageMaker Studio's default storage is EFS, not S3, and this configuration does not ensure encryption at rest for the EFS file system or encryption in transit for notebook traffic. Option C is wrong because enabling SageMaker Studio's default encryption using the AWS managed key for EFS only encrypts data at rest on the EFS file system, but it does not address encryption in transit for notebook traffic, which is required by the policy.

Full explanation →
113
MCQmedium

A company uses SageMaker Clarify to detect bias in a deployed model. The monitoring must run automatically on a schedule. Which SageMaker feature should they use?

A.SageMaker Pipelines
B.SageMaker Experiments
C.SageMaker Data Wrangler
D.SageMaker Model Monitor
AnswerD

Model Monitor can schedule bias detection with Clarify and track drift.

Why this answer

SageMaker Model Monitor can be configured to run bias detection jobs on a schedule using Clarify's bias metrics.

Full explanation →
114
MCQmedium

A startup uses Amazon Lex to build a chatbot for mental health support. They must ensure user conversations are private and not used for model improvement. Which AWS service can help anonymize text data before storage?

A.Amazon Textract
B.AWS Key Management Service (KMS)
C.Amazon Comprehend
D.Amazon Macie
AnswerC

Comprehend's PII detection can be used to redact entities.

Why this answer

Amazon Comprehend offers a built-in feature called PII (Personally Identifiable Information) detection and redaction, which can automatically identify and mask sensitive data such as names, addresses, and health information in text. By using the `DetectPIIEntities` API with redaction, the startup can anonymize user conversations before storing them, ensuring compliance with privacy requirements and preventing data from being used for model improvement.

Exam trap

The trap here is that candidates may confuse data anonymization with data encryption (KMS) or data discovery (Macie), overlooking that Amazon Comprehend provides direct text-level redaction via its PII detection API.

How to eliminate wrong answers

Option A is wrong because Amazon Textract is an OCR service for extracting text from documents (e.g., PDFs, images), not for anonymizing or redacting sensitive data in text. Option B is wrong because AWS KMS manages encryption keys for data at rest or in transit, but it does not perform content-level anonymization or redaction of text. Option D is wrong because Amazon Macie is a data security service that discovers and protects sensitive data in S3 using machine learning, but it operates on stored data and does not provide real-time text anonymization or redaction before storage.

Full explanation →
115
MCQmedium

A company uses Amazon SageMaker to build and deploy machine learning models. The security team has implemented a policy that all SageMaker notebook instances must be launched in a VPC and cannot have direct internet access. However, data scientists need to download open-source datasets from the internet for model training. They also need to install Python packages from PyPI. Currently, the SageMaker notebook instances are in a VPC with no internet gateway. The data scientists cannot download any external data or packages. The VPC has a NAT gateway already configured. What should the company do to allow the necessary internet access while maintaining the security requirement?

A.Attach an internet gateway to the VPC and add a route to the internet gateway in the subnet's route table.
B.Place the notebook instances in a public subnet and configure security group rules to restrict inbound traffic.
C.Create a VPC endpoint for S3 and a VPC endpoint for PyPI, and route traffic through them.
D.Add a route from the private subnets to the NAT gateway for 0.0.0.0/0.
AnswerD

NAT gateway provides outbound internet access for private subnets, allowing downloads without direct inbound access.

Why this answer

The VPC already has a NAT gateway configured, but the private subnets lack a route to it for internet-bound traffic. By adding a route from the private subnets to the NAT gateway for destination 0.0.0.0/0, outbound traffic from the SageMaker notebook instances can reach the internet (e.g., to download datasets from open-source sites and install packages from PyPI) while the instances remain in a private subnet without direct internet access, satisfying the security requirement.

Exam trap

The trap here is that candidates often confuse VPC endpoints (which only work for AWS services like S3 or DynamoDB) with a general internet access solution, forgetting that PyPI and arbitrary websites are not AWS services and thus cannot be reached via VPC endpoints.

How to eliminate wrong answers

Option A is wrong because attaching an internet gateway and adding a route to it would give the notebook instances direct internet access, violating the security policy that they cannot have direct internet access. Option B is wrong because placing notebook instances in a public subnet with a security group restricting inbound traffic still gives them direct internet access via the internet gateway, which breaks the security requirement. Option C is wrong because while VPC endpoints exist for S3, there is no VPC endpoint for PyPI (PyPI is not an AWS service), so this approach cannot route PyPI traffic through a VPC endpoint; additionally, VPC endpoints do not provide general internet access for downloading arbitrary open-source datasets.

Full explanation →
116
MCQhard

A company uses Amazon Bedrock to generate product descriptions. They need to ensure outputs do not contain offensive language. Which service should they integrate to filter content?

A.Amazon Comprehend
B.Amazon Rekognition
C.Bedrock Guardrails
D.AWS WAF
AnswerC

Guardrails offers configurable content filters for safety and compliance.

Why this answer

Amazon Bedrock Guardrails is the correct choice because it is specifically designed to enforce content policies for foundation model outputs, including filtering for offensive language, hate speech, and other harmful content. It integrates directly with Bedrock to apply customizable safety filters and deny topics without requiring additional services or custom code.

Exam trap

The trap here is that candidates often confuse Amazon Comprehend's text analysis capabilities (like sentiment detection) with real-time content filtering, but Comprehend lacks the policy enforcement and integration with Bedrock that Guardrails provides.

How to eliminate wrong answers

Option A is wrong because Amazon Comprehend is a natural language processing (NLP) service for extracting insights like sentiment, entities, and key phrases from text, but it does not provide real-time content filtering or policy enforcement for Bedrock outputs. Option B is wrong because Amazon Rekognition is an image and video analysis service that detects objects, faces, and text in visual media, not a text-based content filter for offensive language. Option D is wrong because AWS WAF is a web application firewall that protects HTTP/HTTPS endpoints from common web exploits like SQL injection and cross-site scripting, not a content moderation filter for LLM-generated text.

Full explanation →
117
Multi-Selecteasy

Which TWO practices help ensure transparency in AI systems? (Choose 2)

Select 2 answers
A.Combine multiple models to obscure decision logic
B.Use model-agnostic explainability tools like SHAP
C.Remove all features except the most predictive ones
D.Provide documentation on model limitations and data sources
E.Use black-box models to protect proprietary algorithms
AnswersB, D

Explainability tools clarify decisions.

Why this answer

Model-agnostic explainability tools like SHAP (SHapley Additive exPlanations) provide post-hoc explanations for any machine learning model by computing feature contributions based on cooperative game theory. This allows stakeholders to understand how each input feature influences a prediction, directly supporting transparency without requiring access to the model's internal structure.

Exam trap

The AIF-C01 exam often tests the misconception that transparency means simplifying the model (e.g., removing features) or hiding logic (e.g., using ensembles or black-box models), when in fact transparency is achieved through explainability tools and thorough documentation of limitations and data sources.

Full explanation →
118
Multi-Selecteasy

A company wants to automatically detect and redact personally identifiable information (PII) from customer support transcripts. Which TWO AWS services can be used together to achieve this? (Choose two.)

Select 2 answers
A.Amazon Comprehend
B.Amazon Rekognition
C.Amazon Transcribe
D.Amazon Textract
E.Amazon Polly
AnswersA, C

Comprehend has a PII detection and redaction feature for text.

Why this answer

Amazon Comprehend is correct because it provides a built-in PII detection and redaction capability via the `DetectPiiEntities` and `ContainsPiiEntities` APIs, which can identify and mask sensitive data such as names, addresses, and credit card numbers in text. Amazon Transcribe is correct because it converts audio customer support calls into text, which is the necessary input format for Comprehend to analyze and redact PII. Together, they form a pipeline: Transcribe generates the transcript, and Comprehend processes it for PII redaction.

Exam trap

The trap here is that candidates often confuse Amazon Rekognition (image/video analysis) or Amazon Textract (document OCR) with text-based PII detection, forgetting that the input source is audio transcripts, not images or scanned documents.

Full explanation →
119
MCQeasy

A company is building a customer support chatbot using Amazon Bedrock. They need to store conversation history for context across sessions. Which AWS service is best suited for this purpose?

A.Amazon S3
B.Amazon DynamoDB
C.Amazon RDS
D.Amazon ElastiCache
AnswerB

DynamoDB provides fast, scalable storage for session state and conversation history.

Why this answer

Amazon DynamoDB is the best choice for storing conversation history because it is a fully managed NoSQL key-value and document database that provides single-digit millisecond latency at any scale. It supports flexible schema, which is ideal for storing variable-length chat sessions, and its Time to Live (TTL) feature can automatically expire old conversations to manage storage costs. DynamoDB also integrates natively with AWS Lambda and Amazon Bedrock for real-time retrieval and update of context across sessions.

Exam trap

The trap here is that candidates often confuse durability with performance, picking Amazon S3 for its low cost or Amazon ElastiCache for its speed, without recognizing that DynamoDB uniquely combines low latency, persistence, and flexible schema for session state management.

How to eliminate wrong answers

Option A is wrong because Amazon S3 is an object storage service designed for large, unstructured data like files and backups, not for low-latency, frequent read/write operations required for real-time conversation history retrieval. Option C is wrong because Amazon RDS is a relational database that requires a fixed schema and is overkill for simple key-value session storage; it also incurs higher operational overhead and latency compared to DynamoDB for this use case. Option D is wrong because Amazon ElastiCache is an in-memory caching service (Redis/Memcached) that is volatile and not designed for durable, persistent storage of conversation history across sessions, making it unsuitable for long-term context retention.

Full explanation →
120
MCQeasy

What is the primary purpose of a model card?

A.To provide a detailed performance benchmark on a single metric
B.To register the model in SageMaker Model Registry
C.To document the model's intended use, performance, and limitations for transparency
D.To store the model's parameters and weights for deployment
AnswerC

Model cards are designed to convey key information about a model to promote responsible use.

Why this answer

Model cards are standardized documents that communicate model details, intended use, performance evaluations, and limitations to users and stakeholders, promoting transparency.

Full explanation →
121
MCQhard

A company uses Amazon SageMaker Clarify to monitor a deployed model for bias. After running an analysis, they find that the model's predictions have a disparate impact on a protected group. What is the MOST appropriate next step?

A.Investigate the root cause of bias, then use techniques such as reweighing training data or applying bias mitigation algorithms before redeploying
B.Modify the SageMaker endpoint to add a random noise to predictions for the protected group
C.Ignore the results because SageMaker Clarify is still in preview
D.Immediately delete the model and retrain from scratch using only data from the protected group
AnswerA

Bias mitigation involves understanding the cause and applying methods like data reweighing or post-processing adjustments.

Why this answer

Discovering bias requires investigation and mitigation. SageMaker Clarify can help identify bias, but mitigation typically involves retraining with balanced data or adjusting model outputs.

Full explanation →
122
MCQeasy

A company uses Amazon SageMaker Model Registry to manage model versions. The security team requires that only approved models can be deployed to production. The ML team creates a new model version and registers it in the Model Registry. When trying to deploy the model to a production endpoint, the deployment fails because the model is not approved. The ML team asks the DevOps team to approve the model. However, the DevOps team cannot see the model in the Model Registry. What is the MOST likely reason?

A.The model artifacts are stored in an S3 bucket that uses SSE-KMS, and the DevOps team does not have KMS decrypt permission.
B.The model version is in a 'Pending' approval status and needs to be promoted by the ML team first.
C.The DevOps team does not have the required IAM permissions to view models in the Model Registry.
D.The Model Registry is in a different AWS account, and the DevOps team does not have cross-account access.
AnswerC

Access to Model Registry is controlled by IAM; the DevOps team likely lacks list/list-approvals permissions.

Why this answer

The most likely reason the DevOps team cannot see the model in the Model Registry is that they lack the required IAM permissions (e.g., `sagemaker:DescribeModelPackage` or `sagemaker:ListModelPackages`) to view models. Without these permissions, the model is invisible to them in the SageMaker console or API, even though it exists. The deployment failure due to approval status is a separate symptom; the core issue here is visibility, not approval workflow.

Exam trap

The trap here is that candidates confuse the approval status (Pending) with visibility, assuming a model in 'Pending' status is hidden, when in fact the approval status only controls deployment, not the ability to see the model in the registry.

How to eliminate wrong answers

Option A is wrong because SSE-KMS permissions affect the ability to decrypt model artifacts when deploying or downloading them, not the ability to see the model in the Model Registry. Option B is wrong because a 'Pending' approval status would prevent deployment but would not hide the model from the DevOps team; they would still be able to see it in the registry. Option D is wrong because while cross-account access could cause visibility issues, the question states the Model Registry is used by the company, implying a single-account setup, and there is no mention of multiple accounts.

Full explanation →
123
MCQmedium

A company deployed a question-answering system using Amazon Bedrock with a knowledge base (RAG). Users report that the model often hallucinates facts not in the knowledge base. What is the most effective way to reduce hallucinations?

A.Reduce the maximum context length to limit model input
B.Fine-tune the foundation model on a large general corpus
C.Improve the relevance of retrieved documents by refining the retrieval strategy
D.Increase the chunk size of documents in the knowledge base
AnswerC

Better retrieval ensures only pertinent information is provided, reducing the chance of hallucination.

Why this answer

Hallucinations in RAG systems often stem from the model receiving irrelevant or low-quality retrieved documents, which forces it to rely on its parametric knowledge rather than the provided context. By refining the retrieval strategy—such as improving embedding quality, adjusting chunk overlap, or using hybrid search—the system ensures the foundation model has the most relevant information to ground its answers, directly reducing the likelihood of fabricating facts.

Exam trap

A common misconception is that hallucinations are primarily a model training issue (fine-tuning or context length) rather than a retrieval quality issue in RAG systems, leading candidates to overlook the critical role of the retriever in grounding responses.

How to eliminate wrong answers

Option A is wrong because reducing the maximum context length limits the amount of retrieved context the model can use, which actually increases the risk of hallucinations by forcing the model to rely more on its own training data rather than the knowledge base. Option B is wrong because fine-tuning on a large general corpus would further embed general knowledge into the model, potentially exacerbating hallucinations when the model defaults to its training data instead of the knowledge base; fine-tuning is not a targeted fix for retrieval quality. Option D is wrong because increasing chunk size can lead to chunks that contain irrelevant or noisy information, reducing the precision of retrieval and potentially introducing more irrelevant context that confuses the model, rather than improving answer accuracy.

Full explanation →
124
MCQeasy

A company is building a generative AI application for code generation. They want to minimize costs while maintaining acceptable performance for their workload, which has periodic spikes in demand. Which approach would be MOST cost-effective?

A.Right-size model selection: use a smaller model for simple tasks and a larger model only when needed
B.Always use the largest available foundation model for all requests
C.Use model caching to store responses for repeated prompts
D.Use batch inference for all requests
AnswerA

Right-sizing reduces cost by matching model capability to task complexity.

Why this answer

Using a smaller, faster model for most requests and a larger model only for complex tasks balances cost and performance, especially with demand spikes.

Full explanation →
125
MCQmedium

A financial institution is building a model to approve loan applications. They must comply with the EU AI Act, which classifies credit scoring as a high-risk AI system. Which requirement is the MOST likely to apply under the EU AI Act?

A.The model must be trained exclusively on data from within the EU
B.The model must achieve a minimum accuracy of 95%
C.The system must be able to be overridden or stopped by a human
D.The system must be explainable using SHAP values
AnswerC

High-risk systems require human oversight mechanisms, including the ability to override or stop the system.

Why this answer

High-risk AI systems under the EU AI Act require human oversight to ensure that decisions can be reviewed and overridden. Transparency and bias reporting are also requirements, but human oversight is specifically mandated to allow intervention.

Full explanation →
126
MCQhard

A financial services company uses Amazon Bedrock to generate investment advice. They have configured a guardrail to deny any harmful content. However, a user prompt 'Tell me how to commit fraud' was not blocked. What is the most likely cause?

A.The guardrail's sensitivity threshold is set too low
B.The guardrail's content policy does not include the 'Fraud' category
C.The prompt was sent to a different Bedrock model that does not support guardrails
D.The guardrail was configured with a word filter but not a content filter
AnswerB

Content policies are configurable; if 'Fraud' is not selected, the guardrail will not block prompts related to fraud.

Why this answer

Bedrock Guardrails rely on content policies that detect harmful content based on categories like fraud. If the guardrail template does not include the 'Fraud' category, such prompts may pass through. The guardrail's configuration must explicitly include the relevant harm category.

Full explanation →
127
MCQmedium

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?

A.AWS CodePipeline and Amazon CodeGuru
B.Amazon CloudWatch Events and AWS CloudTrail
C.AWS Organizations and AWS Artifact
D.AWS Config custom rules and AWS IAM policies
AnswerD

Config can evaluate tags on models, IAM can restrict deployment to roles that can only deploy approved models.

Why this answer

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.

Exam trap

The trap here is that candidates often confuse monitoring/auditing services (like CloudTrail and CloudWatch) with enforcement mechanisms, failing to recognize that only AWS Config rules combined with IAM policies can actively prevent or flag non-compliant deployments.

How to eliminate wrong answers

Option A is wrong because AWS CodePipeline is a CI/CD service for automating build and deploy pipelines, and Amazon CodeGuru provides code reviews and profiling, neither of which can enforce governance over which specific ML models are deployed to SageMaker. Option B is wrong because Amazon CloudWatch Events (now Amazon EventBridge) and AWS CloudTrail are monitoring and auditing services that record API calls and trigger events, but they cannot prevent or enforce deployment of only approved models. Option C is wrong because AWS Organizations manages multi-account governance and service control policies, and AWS Artifact provides compliance reports, but neither can directly evaluate or restrict SageMaker model deployment approvals.

Full explanation →
128
MCQhard

A security engineer creates the above IAM policy to allow a user to invoke an Amazon Bedrock model. However, invocation fails. What is the issue?

A.The action should be "bedrock:InvokeModelWithResponseStream".
B.The resource ARN is missing the account ID.
C.The ARN should use "foundation-model" instead of "model".
D.The statement is missing a condition for the model ID.
AnswerC

The resource type for foundation models is 'foundation-model', not 'model'.

Why this answer

The IAM policy's resource ARN incorrectly uses 'model' in the path, but Amazon Bedrock requires 'foundation-model' to reference foundation models. The correct ARN format for invoking a Bedrock foundation model is 'arn:aws:bedrock:region::foundation-model/model-id'. Using 'model' instead of 'foundation-model' causes the policy to not match any valid Bedrock resource, resulting in an invocation failure.

Exam trap

AWS often tests the distinction between 'model' and 'foundation-model' in Bedrock ARNs, as candidates may assume all Bedrock models use the same resource type, overlooking that foundation models require a specific path.

How to eliminate wrong answers

Option A is wrong because 'bedrock:InvokeModelWithResponseStream' is a separate action for streaming responses, but the standard 'bedrock:InvokeModel' action is sufficient for non-streaming invocation; the failure is not due to the action name. Option B is wrong because the resource ARN for Bedrock foundation models does not require an account ID; the ARN format uses a double colon (::) in the account ID position, which is correct for service-owned resources. Option D is wrong because a condition for the model ID is optional and not required for invocation; the primary issue is the incorrect resource type in the ARN.

Full explanation →
129
MCQmedium

A company uses Amazon Bedrock to build an AI assistant. They need to restrict the model from generating responses about competitors. Which Bedrock feature should they configure?

A.Word filters
B.Denied topics
C.PII redaction
D.Content filters
AnswerB

Denied topics in Guardrails explicitly prevent the model from discussing prohibited subjects.

Why this answer

Bedrock Guardrails allows you to define topic restrictions that block certain topics from being discussed. Other options are for content filtering, PII redaction, or grounding.

Full explanation →
130
MCQmedium

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?

A.Amazon Personalize
B.Amazon Rekognition
C.Amazon Comprehend
D.Amazon Forecast
AnswerA

Personalize is designed for real-time personalization and recommendations.

Why this answer

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.

Exam trap

The trap here is that candidates may confuse Amazon Personalize with Amazon Forecast, as both involve 'predicting' something, but Forecast is for time-series numeric predictions (e.g., sales volume) while Personalize is for user-specific item recommendations.

How to eliminate wrong answers

Option B (Amazon Rekognition) is wrong because it is a computer vision service for analyzing images and videos, not for generating product recommendations. Option C (Amazon Comprehend) is wrong because it is a natural language processing (NLP) service for extracting insights from text, such as sentiment or entities, not for building recommendation engines. Option D (Amazon Forecast) is wrong because it is a time-series forecasting service for predicting future metrics like demand or sales, not for personalizing recommendations based on user behavior.

Full explanation →
131
MCQmedium

A company uses Amazon Bedrock Agents to automate a multi-step data processing workflow. The agent needs to call an external API to enrich customer records. How should the developer expose this API to the agent?

A.By embedding the API directly in the agent's prompt instructions
B.By creating a custom model that has been fine-tuned to call the API
C.By defining an action group with the API specification and a Lambda function
D.By configuring a Bedrock Knowledge Base with API documentation
AnswerC

Action groups allow the agent to call external APIs via defined schemas and Lambda functions.

Why this answer

Action groups define the tools an agent can invoke. Each action group contains Lambda functions or API schemas that the agent calls during execution.

Full explanation →
132
MCQhard

A model trained to predict credit risk shows that applicants from a certain zip code are disproportionately rejected, even though income and credit history are comparable. Which type of bias is MOST likely present?

A.Measurement bias
B.Historical bias
C.Representation bias
D.Aggregation bias
AnswerC

Representation bias occurs when the training data does not adequately represent the population, leading to poor generalizability for underrepresented groups. Here, the model likely learned a spurious correlation with zip code due to imbalanced representation.

Why this answer

Representation bias occurs when the training data over- or under-represents certain groups, leading the model to learn spurious correlations like zip code with creditworthiness.

Full explanation →
133
MCQhard

A financial services company uses Amazon SageMaker to train models with sensitive customer data. They must ensure that no data leaves a specific AWS Region due to data residency regulations. The training data is in S3. Which architecture meets this requirement while minimizing data transfer?

A.Place SageMaker training job in a private subnet with a NAT gateway and route traffic through the internet
B.Configure S3 Transfer Acceleration and use a public SageMaker training job
C.Use AWS Glue to copy data to an EBS volume attached to the training instance
D.Use S3 VPC endpoints and place SageMaker training job in a private subnet with no internet access
AnswerD

VPC endpoints keep S3 traffic within AWS network and same region; no internet access ensures residency.

Why this answer

Using a VPC with S3 VPC endpoints ensures data stays within the AWS network and does not traverse the internet. Data remains in the same Region because S3 endpoints are Regional.

Full explanation →
134
MCQeasy

Which AWS service provides managed foundation models from providers like Anthropic, Meta, and Stability AI through a single API?

A.AWS Lambda
B.Amazon Bedrock
C.Amazon SageMaker
D.Amazon Rekognition
AnswerB

Bedrock provides access to foundation models from multiple providers through a single API.

Why this answer

Amazon Bedrock is a fully managed service that provides access to foundation models (FMs) from leading AI companies such as Anthropic (Claude), Meta (Llama), and Stability AI (Stable Diffusion) through a single, unified API. This allows developers to integrate and experiment with multiple FMs without managing underlying infrastructure or dealing with separate provider endpoints.

Exam trap

The trap here is that candidates may confuse Amazon Bedrock with Amazon SageMaker, thinking SageMaker also provides managed foundation models, but SageMaker is primarily for custom model training and deployment, not for consuming pre-built third-party FMs via a single API.

How to eliminate wrong answers

Option A is wrong because AWS Lambda is a serverless compute service for running code in response to events, not a managed foundation model service; it does not provide access to pre-trained models via a single API. Option C is wrong because Amazon SageMaker is a machine learning platform for building, training, and deploying custom models, not a managed service for consuming third-party foundation models through a unified API. Option D is wrong because Amazon Rekognition is a specialized computer vision service for image and video analysis, not a general-purpose foundation model hub that includes models from Anthropic, Meta, or Stability AI.

Full explanation →
135
Multi-Selectmedium

A data scientist is preparing data for a classification model. The dataset contains missing values in several features. Which TWO approaches are appropriate for handling missing data? (Select TWO.)

Select 2 answers
A.Remove rows with any missing values
B.Set missing values to zero
C.Ignore missing values during training
D.Replace missing values with -1
E.Impute missing values with the median of the feature
AnswersA, E

Listwise deletion is valid if the missingness is random and the sample size remains sufficient.

Why this answer

Imputation fills missing values with estimated values (e.g., mean, median, mode), allowing the model to use all samples. Removing rows with missing values is a simple deletion approach.

Full explanation →
136
MCQmedium

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?

A.SageMaker Ground Truth
B.SageMaker Clarify
C.SageMaker Automatic Model Tuning
D.SageMaker Model Monitor
AnswerB

Clarify provides feature importance, SHAP values, and bias metrics for model explainability.

Why this answer

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.

Exam trap

The trap here is confusing monitoring (Model Monitor) with explainability (Clarify), as both relate to model governance but serve fundamentally different purposes—monitoring tracks performance over time, while Clarify explains individual predictions.

How to eliminate wrong answers

Option A is wrong because SageMaker Ground Truth is a data labeling service for creating training datasets, not for explaining model predictions. Option C is wrong because SageMaker Automatic Model Tuning (hyperparameter optimization) adjusts model parameters to improve performance, but does not provide explainability or feature attribution. Option D is wrong because SageMaker Model Monitor detects data drift and model quality degradation over time, but does not generate explanations for individual predictions.

Full explanation →
137
Multi-Selecteasy

Which TWO techniques can reduce the cost of running a fine-tuned foundation model on Amazon SageMaker? (Choose TWO.)

Select 2 answers
A.Implement structured pruning to remove less important model parameters.
B.Use larger instance types with more GPUs to speed up inference.
C.Apply model quantization to reduce precision from FP32 to FP16 or INT8.
D.Store the model parameters in FP32 to maintain accuracy during inference.
E.Increase the number of training epochs to achieve higher accuracy.
AnswersA, C

Pruning creates a smaller model that is cheaper to run.

Why this answer

Structured pruning reduces the number of parameters in the model by removing entire neurons, channels, or layers that contribute little to the output. This directly shrinks the model size and computational requirements, leading to lower memory usage and faster inference on SageMaker, which reduces cost.

Exam trap

AWS often tests the distinction between techniques that reduce inference cost (pruning, quantization) versus those that improve training speed or accuracy, leading candidates to mistakenly select options that increase resource usage or are irrelevant to inference cost.

Full explanation →
138
MCQeasy

An organization wants to document key information about their machine learning model, including intended use, performance metrics, training data, and ethical considerations. Which tool or practice should they adopt?

A.SageMaker Model Registry
B.Data sheets
C.Model cards
D.AWS CloudTrail logs
AnswerC

Model cards provide a standardized format for documenting model details, including intended use, performance, and ethical considerations.

Why this answer

Model cards are structured documentation templates that capture relevant model details, promoting transparency and responsible AI practices. They are recommended by responsible AI guidelines.

Full explanation →
139
Multi-Selectmedium

A company needs to select a vector store for their Amazon Bedrock Knowledge Base. Which TWO options are supported as vector stores? (Choose TWO.)

Select 2 answers
A.Amazon Aurora pgvector
B.Amazon OpenSearch Serverless
C.Amazon Redshift
D.Amazon RDS for MySQL
E.Amazon DynamoDB
AnswersA, B

Aurora with pgvector extension supports vector storage and search.

Why this answer

Amazon Bedrock Knowledge Bases support Amazon OpenSearch Serverless, Aurora pgvector, Pinecone, and MongoDB Atlas. DynamoDB and Redshift are not vector stores.

Full explanation →
140
MCQmedium

A company wants to build a customer service chatbot that answers questions about their internal policy documents. The documents are updated monthly, and the team cannot afford to retrain a model each time. Which approach is MOST appropriate?

A.Train a custom model from scratch on the policy documents each month
B.Fine-tune a base LLM on the policy documents monthly
C.Use Retrieval-Augmented Generation (RAG) with the policy documents indexed in a vector store
D.Use a larger foundation model with a longer context window and paste all documents into each prompt
AnswerC

RAG retrieves relevant document chunks at query time, ensuring the chatbot always answers from the latest uploaded documents without retraining.

Why this answer

RAG allows the LLM to retrieve relevant document sections at inference time, so knowledge stays current without retraining. The other options either require expensive retraining or lack document grounding.

Full explanation →
141
MCQhard

A healthcare organization is developing a clinical decision support system using Amazon Bedrock with a large language model (LLM) to analyze patient symptoms and suggest potential diagnoses. The system must comply with HIPAA and internal responsible AI guidelines. During testing, the model occasionally generates diagnoses that are inconsistent with established medical guidelines and shows a tendency to recommend more aggressive treatments for patients from certain demographic groups. The team has already implemented data encryption, access controls, and basic content filtering. They need to further reduce biased and unsafe outputs without delaying the deployment timeline. What should the team do next?

A.Increase the logging of all model inputs and outputs to Amazon CloudWatch and set up alarms for any mentions of protected attributes.
B.Replace the current LLM with a different pre-trained model that has been benchmarked for lower bias on medical datasets.
C.Fine-tune the model using a curated dataset of anonymized patient records that is balanced across demographic groups and aligned with clinical guidelines.
D.Apply stronger content filtering rules using Amazon Comprehend Medical to block any diagnosis that contains demographic-related terms.
AnswerC

Fine-tuning on a balanced, guideline-aligned dataset reduces both bias and inaccuracy by teaching the model correct patterns.

Why this answer

Fine-tuning the model with a balanced, curated dataset directly addresses both the bias and clinical accuracy issues at the model level, which is the most effective approach for reducing biased and unsafe outputs without delaying deployment. This method adjusts the model's internal weights to align with established medical guidelines and demographic fairness, rather than relying on post-processing filters or logging that do not fix the root cause. Since the team has already implemented basic content filtering, fine-tuning provides a targeted, efficient solution that can be completed within a reasonable timeline.

Exam trap

The trap here is that candidates may confuse monitoring and logging (Option A) with actual bias mitigation, or assume that a different pre-trained model (Option B) will inherently solve domain-specific bias without requiring additional fine-tuning or validation.

How to eliminate wrong answers

Option A is wrong because increasing logging and setting alarms for protected attributes only monitors for bias after it occurs, but does not prevent or reduce biased or unsafe outputs; it adds operational overhead without addressing the model's behavior. Option B is wrong because replacing the current LLM with a different pre-trained model introduces significant risk of deployment delays due to re-evaluation, integration, and compliance validation, and does not guarantee lower bias on the specific medical domain without further customization. Option D is wrong because applying stronger content filtering with Amazon Comprehend Medical to block diagnoses containing demographic terms is a blunt, post-processing approach that can suppress legitimate clinical information and still allow biased patterns that do not explicitly mention protected attributes, failing to address the underlying model bias.

Full explanation →
142
MCQeasy

A data scientist needs to restrict access to a SageMaker notebook instance to only the corporate network. Which configuration should they use?

A.Enable multi-factor authentication for the notebook
B.Use an IAM policy to allow only corporate users
C.Place the notebook instance in a VPC and configure security groups to allow only corporate IP ranges
D.Use a SageMaker lifecycle configuration to block external IPs
AnswerC

VPC placement and security groups enforce network-level restrictions.

Why this answer

VPC isolation places the notebook instance inside a VPC, and security groups can restrict inbound traffic to the corporate network IP range.

Full explanation →
143
MCQhard

A machine learning engineer notices that the training loss decreases steadily, but the validation loss starts increasing after a few epochs. Which of the following is the MOST likely cause?

A.Learning rate too low
B.Overfitting
C.Underfitting
D.Data leakage from validation set into training set
AnswerB

Classic overfitting: model memorizes training data and fails to generalize to validation data.

Why this answer

The scenario describes training loss decreasing while validation loss increases after a few epochs, which is the classic signature of overfitting. The model is memorizing the training data (including noise) rather than learning generalizable patterns, causing it to perform poorly on unseen validation data.

Exam trap

AWS AI Practitioner often tests the distinction between overfitting and underfitting by describing loss curves; the trap here is that candidates may confuse a rising validation loss with a learning rate issue or data leakage, but the steady decrease in training loss rules out underfitting and points directly to overfitting.

How to eliminate wrong answers

Option A is wrong because a learning rate that is too low would cause both training and validation loss to decrease very slowly or plateau, not cause validation loss to increase after initially decreasing. Option C is wrong because underfitting occurs when the model is too simple to capture patterns in the data, resulting in both training and validation loss remaining high and not decreasing steadily. Option D is wrong because data leakage from the validation set into the training set would artificially inflate training performance and likely cause both losses to be low and correlated, not a divergence where validation loss increases.

Full explanation →
144
MCQeasy

A company wants to detect sensitive data such as PII in their training datasets stored in S3 before using them for model training. Which AWS service should they use?

A.Amazon Macie
B.Amazon Inspector
C.AWS Shield
D.Amazon GuardDuty
AnswerA

Macie automatically discovers sensitive data in S3 using machine learning.

Why this answer

Amazon Macie uses machine learning to discover and protect sensitive data in S3, including PII.

Full explanation →
145
MCQeasy

A company uses Amazon Rekognition to analyze images stored in an S3 bucket. The security team requires that all image analysis be logged to AWS CloudTrail for auditing. What is the minimum configuration needed to meet this requirement?

A.Enable Rekognition logging in the AWS Management Console
B.Enable CloudTrail management events for the S3 bucket
C.Enable S3 server access logs on the bucket
D.Enable CloudTrail data events for the S3 bucket to capture GetObject API calls
AnswerD

Data events capture object-level operations; Rekognition calls GetObject when reading images.

Why this answer

CloudTrail data events capture S3 object-level API operations such as GetObject, which is the API call made by Amazon Rekognition when it retrieves images from the S3 bucket for analysis. By enabling data events for the S3 bucket, every GetObject request is logged to CloudTrail, providing the audit trail the security team requires. Management events alone do not capture object-level operations, and S3 server access logs are not integrated with CloudTrail for auditing.

Exam trap

The trap here is that candidates often confuse management events with data events, assuming that enabling CloudTrail for the S3 bucket automatically captures all API calls, when in fact management events only cover control-plane operations and not the object-level GetObject calls made by Rekognition.

How to eliminate wrong answers

Option A is wrong because Amazon Rekognition does not have a separate logging configuration in the AWS Management Console; its API calls are logged via CloudTrail when data events are enabled for the relevant S3 bucket. Option B is wrong because CloudTrail management events capture control-plane operations (e.g., bucket creation, policy changes) but do not capture data-plane operations like GetObject, which is the specific API call used by Rekognition to read images. Option C is wrong because S3 server access logs provide detailed records of requests made to the bucket, but they are not part of CloudTrail and do not satisfy the requirement for auditing via CloudTrail; they are a separate logging mechanism.

Full explanation →
146
MCQmedium

A healthcare organization uses an ML model to predict patient readmission risk. To comply with regulations, they need to explain individual predictions to clinicians. Which explainability technique provides local, model-agnostic explanations that are computationally efficient?

A.Partial dependence plots
B.Amazon SageMaker Autopilot
C.Global feature importance from a random forest
D.LIME (Local Interpretable Model-agnostic Explanations)
AnswerD

LIME provides local explanations for individual predictions, is model-agnostic, and is computationally efficient compared to SHAP.

Why this answer

LIME (Local Interpretable Model-agnostic Explanations) is a model-agnostic method that approximates the model locally to provide explanations for individual predictions. SHAP is also local and model-agnostic but can be computationally intensive.

Full explanation →
147
MCQmedium

A company wants to build a model to forecast monthly sales. The data is a time series with trend and seasonality. Which SageMaker algorithm is most appropriate?

A.XGBoost
B.K-Means
C.Linear Learner
D.DeepAR
AnswerD

DeepAR is a built-in SageMaker algorithm specifically for time series forecasting with seasonality and trend.

Why this answer

DeepAR is the most appropriate algorithm because it is specifically designed for time series forecasting, handling both trend and seasonality through autoregressive recurrent neural networks. It learns from multiple related time series and produces probabilistic forecasts, making it ideal for monthly sales prediction.

Exam trap

The trap here is that candidates often choose XGBoost or Linear Learner because they are familiar with regression tasks, but fail to recognize that time series forecasting requires algorithms that explicitly model temporal dependencies and seasonality, which DeepAR is built for.

How to eliminate wrong answers

Option A is wrong because XGBoost is a gradient boosting algorithm for tabular data, not designed to capture temporal dependencies or seasonality in time series without extensive feature engineering. Option B is wrong because K-Means is an unsupervised clustering algorithm that groups data points by similarity, with no capability for forecasting sequential data. Option C is wrong because Linear Learner is a linear regression model that assumes independence of observations and cannot model complex time series patterns like seasonality or long-term trends.

Full explanation →
148
MCQmedium

A startup is building an AI-powered code assistant using a large language model (LLM). They want to ensure the model generates syntactically correct code and avoids security vulnerabilities. Which technique should they prioritize?

A.Augment prompts with few-shot examples of secure coding practices and unit tests
B.Deploy the model with max tokens set to 4096
C.Fine-tune the model on a large corpus of open-source code
D.Use chain-of-thought prompting to explain reasoning before code generation
AnswerA

Providing examples of secure code and expected test results helps ground the model's output in desired patterns.

Why this answer

Contextual grounding by providing code examples and security guidelines in the prompt (prompt engineering) helps guide the model to produce safe and correct code. Fine-tuning on secure codebases would also help but is more resource-intensive; prompt engineering is a quicker first step.

Full explanation →
149
MCQhard

A company is deploying a machine learning model for real-time fraud detection. The model must make predictions with latency under 10 milliseconds. The data scientist trained a gradient boosting model that achieves high accuracy but has inference latency of 50 milliseconds. The team has access to a larger instance type with more CPU cores. Which approach should the data scientist take to reduce inference latency while maintaining accuracy?

A.Switch to batch inference and run predictions every 100 milliseconds.
B.Deploy the model on a larger instance with more CPU cores.
C.Reduce the maximum tree depth and retrain the model.
D.Apply post-training pruning to remove redundant trees.
AnswerB

More CPU cores allow parallel computation, reducing inference latency without changing the model.

Why this answer

Increasing the number of CPU cores allows the gradient boosting model to parallelize tree evaluation across multiple cores, reducing inference latency. Since the model is already trained and accurate, this hardware scaling directly addresses the 50 ms bottleneck without altering the model's structure or accuracy.

Exam trap

AWS often tests the misconception that model optimization (pruning or depth reduction) is the only way to reduce latency, ignoring that hardware scaling (more CPU cores) can meet latency requirements without sacrificing accuracy.

How to eliminate wrong answers

Option A is wrong because switching to batch inference every 100 ms violates the real-time requirement of under 10 ms latency per prediction; it introduces a fixed delay that exceeds the threshold. Option C is wrong because reducing maximum tree depth reduces model complexity, which can lower accuracy and may not guarantee latency under 10 ms if the number of trees remains high. Option D is wrong because post-training pruning removes trees, which reduces model size but can degrade accuracy, and the latency improvement may be insufficient if the remaining trees still require sequential evaluation on limited cores.

Full explanation →
150
MCQeasy

A data scientist sets up a Model Monitoring schedule for data quality. What is a potential security issue with this configuration?

A.The monitoring job uses a single role for both training and monitoring, violating least privilege
B.The schedule runs every hour, which may generate too many logs
C.The monitoring job uses the same endpoint as the production model
D.The output is stored in an S3 bucket with no encryption
AnswerA

Best practice is to have separate roles for different tasks to limit permissions.

Why this answer

Using a single AWS Identity and Access Management (IAM) role for both the training job and the monitoring job violates the principle of least privilege. The training role typically requires broader permissions (e.g., access to training datasets, SageMaker full access), while the monitoring role only needs read-only access to the endpoint and write access to the monitoring output location. Sharing a single role increases the blast radius if the monitoring job is compromised, as an attacker could leverage the elevated training permissions to modify or exfiltrate data.

Exam trap

AWS often tests the principle of least privilege by presenting a seemingly harmless configuration (like a shared role) and distracting candidates with operational or encryption concerns that are less directly tied to the security of the monitoring schedule itself.

How to eliminate wrong answers

Option B is wrong because running a monitoring schedule every hour is a standard practice for data quality checks and does not inherently create a security issue; excessive logging is an operational concern, not a security vulnerability. Option C is wrong because using the same endpoint for monitoring and production is expected — Model Monitoring captures inference requests from the production endpoint to analyze data drift or quality; sharing the endpoint does not introduce a security issue. Option D is wrong because storing output in an unencrypted S3 bucket is a data-at-rest compliance risk, but the question specifically asks for a 'potential security issue' with the monitoring schedule configuration, and the most direct security flaw is the IAM role misconfiguration, not the encryption setting.

Full explanation →
Page 1

Page 2 of 9

Page 3

All pages

123456789

Practice AIF-C01 by domain

Target a specific domain to shore up weak areas.

Applications of Foundation ModelsSecurity, Compliance, and Governance for AI SolutionsFundamentals of AI and MLFundamentals of Generative AIGuidelines for Responsible AI
See all domains with question counts →