The correct answer is A, as the most effective way to ensure fairness in AI systems is to audit and diversify training data. AWS Responsible AI documentation stresses that bias originates primarily from unbalanced or incomplete datasets. By including samples from all demographic groups, the model learns patterns that generalize fairly across populations. AWS SageMaker Clarify supports data bias detection and offers fairness metrics like demographic parity and equal opportunity difference. Deeper models (option B) or secrecy (option C) do not mitigate bias and may worsen transparency. Static test datasets (option D) fail to capture evolving data distributions. Therefore, ongoing bias reviews and diverse datasets form the foundation of responsible generative AI for financial use cases.

Referenced AWS AI/ML Documents and Study Guides:

AWS Responsible AI Whitepaper – Fairness and Bias Mitigation

Amazon SageMaker Clarify Documentation – Bias Detection

AWS documentation identifies hallucinations and inaccuracies as a key challenge when deploying generative AI models in production environments. Hallucinations occur when a model generates responses that are plausible-sounding but factually incorrect, unsupported, or misleading.

Generative AI models are probabilistic by nature and do not have an inherent understanding of truth. AWS emphasizes that these models generate outputs based on patterns learned from training data, which can lead to confident but incorrect responses, especially when prompts lack sufficient context or when the model is asked about information outside its training scope.

The other options do not represent disadvantages. High accuracy and reliability are desired outcomes, not limitations. Deterministic behavior is not typical of generative models and is not a disadvantage. Negligible computational requirements are incorrect, as generative models typically require significant compute resources.

AWS recommends mitigation strategies such as Retrieval Augmented Generation, human review, prompt engineering, and output validation to reduce hallucinations. Nevertheless, hallucinations remain a known risk, making this option the correct answer.

Retrieval Augmented Generation (RAG) combines information retrieval with foundation model inference to generate contextually relevant answers.

Instead of requiring extensive labeled training datasets, RAG retrieves relevant information from existing enterprise data sources and supplies that information to a foundation model as contextual input.

For this scenario, the product database already contains the information needed to answer customer questions. A RAG architecture can retrieve relevant product descriptions during inference without retraining the foundation model.

Amazon Bedrock Knowledge Bases offers managed retrieval capabilities that simplify implementing this architecture.

Fine-tuning requires additional model customization and maintenance. Deploying separate models increases operational complexity, while training a foundation model from scratch requires substantial computational resources and expertise.

RAG therefore provides the most practical solution with comparatively low implementation effort, particularly when product information changes frequently.

The ML research team needs a mechanism to audit custom ML models while sharing model artifacts with other teams. Amazon SageMaker Model Cards provide a structured way todocument model details, including intended uses, training data, and inference performance, making them ideal for auditing and ensuring transparency when publishing models.

Exact Extract from AWS AI Documents:

From the Amazon SageMaker Developer Guide:

" Amazon SageMaker Model Cards enable you to document critical details about your machine learning models, such as intended uses, training data, evaluation metrics, and inference details. Model Cards support auditing by providing a centralized record that can be reviewed by teams to understand model behavior and limitations. "

(Source: Amazon SageMaker Developer Guide, SageMaker Model Cards)

Detailed Explanation:

Option A: Create documents with the relevant information. Store the documents in Amazon S3.While storing documents in S3 is feasible, it lacks the structured format and integration with SageMaker that Model Cards provide, making it less suitable for auditing purposes.

Option B: Use AWS AI Service Cards for transparency and understanding models.AWS AI Service Cards are not a standard feature in AWS documentation. This option appears to be a distractor and is not a valid solution.

Option C: Create Amazon SageMaker Model Cards with Intended uses and training and inference details.This is the correct answer. SageMaker Model Cards are specifically designed to document model details for auditing, transparency, and collaboration, meeting the team’s requirements.

Option D: Create model training scripts. Commit the model training scripts to a Git repository.Sharing training scripts in a Git repository provides access to code but does not offer a structured auditing mechanism for model details like intended uses or inference performance.

Retrieval-Augmented Generation (RAG) integrates external knowledge sources (databases, vector stores, document repositories) with LLMs, enabling them to generate contextually accurate and up-to-date responses without retraining.

B is incorrect: RAG does not speed up training; it improves inference results.

C is incorrect: speech recognition is not an RAG use case.

D is incorrect: computer vision augmentation is unrelated to RAG.

???? Reference:

AWS Documentation – Knowledge Bases for RAG in Amazon Bedrock

Fine-tuning involves training an already pre-trained FM on a smaller, labeled dataset for task specialization.

Data selection is about curating training data.

Pre-training is the initial training phase on massive datasets.

Evaluation happens after training, not during.

???? Reference:

AWS Documentation – Fine-tuning in Amazon Bedrock

AWS responsible AI guidance emphasizes that predictive accuracy is not the only consideration when selecting an AI/ML model. In regulated applications such as financial lending, organizations may have explicit requirements to explain why a decision was made. This creates a tradeoff between predictive performance and model interpretability.

AWS Prescriptive Guidance states regarding tree models: “For single tree models, the split variables and leaf values provide an immediately explainable model.” This is particularly important for a loan-approval scenario because stakeholders may need to determine which applicant characteristics influenced a rejection or approval.

AWS also explains that model interpretability can be a required business outcome in regulated industries, specifically including finance. Interpretability can help organizations justify important decisions, support fairness, satisfy auditing requirements, build trust, and investigate problematic predictions.

Deep neural networks can achieve excellent predictive performance, but their large numbers of parameters and nonlinear interactions generally make their internal decision-making more difficult for humans to interpret directly. Explainability techniques such as SHAP can provide feature attribution for complex models, but this is different from using a model whose decision structure is inherently straightforward.

A decision tree represents decisions through observable splits and branches. For example, an auditor can examine which features and thresholds produced a particular path through the tree. That makes the decision mechanism inherently easier to communicate and inspect.

Option A incorrectly assumes predictive accuracy implies explainability. Option C incorrectly separates complexity from interpretability. Option D is incorrect because different model classes can vary substantially in transparency even when solving the same business problem.

Therefore, where regulatory explainability is a mandatory requirement, the organization should explicitly consider the accuracy-versus-interpretability tradeoff, making B the correct answer .

===========

The verified answer is A. AWS PrivateLink. The requirement is that API calls between generative AI applications and foundation models must not travel across the public internet. AWS documentation states that AWS PrivateLink can create a private connection between a VPC and Amazon Bedrock. With this configuration, applications can access Amazon Bedrock as if it were inside the VPC, without requiring an internet gateway, NAT device, VPN connection, or Direct Connect connection. AWS also states that instances in the VPC do not need public IP addresses to access Amazon Bedrock. This exactly satisfies the requirement for private API communication to foundation models.

Amazon Q is incorrect because Amazon Q is a generative AI assistant service for business, developer, and data use cases. It is not the network service used to privately route API calls between applications and FMs.

Amazon CloudFront is incorrect because CloudFront is a content delivery network. It caches and delivers content through edge locations, but it is not the correct service for private VPC-to-AWS-service connectivity that avoids the public internet.

AWS CloudTrail is incorrect because CloudTrail records account activity and API events for governance, audit, and security analysis. It helps track what happened, but it does not provide private network connectivity.

The key phrase is “must not travel across the public internet.” In AWS architecture, that requirement points to private connectivity using VPC endpoints powered by AWS PrivateLink. Therefore, the correct service is AWS PrivateLink.

The verified answer is B. Amazon SageMaker Clarify. The model is used for customer loan applications, where explainability, transparency, and human-centered decision support are critical. AWS documentation states that SageMaker Clarify provides tools to help explain how machine learning models make predictions. These tools help ML developers, modelers, and internal stakeholders understand model characteristics before deployment and debug predictions after deployment.

SageMaker Clarify is specifically aligned with explainable AI because it provides feature attribution methods, including SHAP-based explanations. AWS documentation explains that Clarify can generate reports that contribute to model governance documentation and can help describe how a model works, what features influence its decisions, and how explanations are generated.

Amazon SageMaker Model Registry is incorrect because it manages model versions, approval status, and deployment tracking. It supports governance workflows, but it does not explain why a model made a decision.

Amazon SageMaker Pipelines is incorrect because it automates and orchestrates ML workflows such as training, evaluation, and deployment. It does not provide the core explainability function required here.

Amazon SageMaker Feature Store is incorrect because it stores, shares, and serves ML features for training and inference. It improves feature reuse and consistency, but it does not explain model decisions.

For a financial loan application model, stakeholders and regulators need to understand which factors influenced predictions. SageMaker Clarify is the AWS feature designed for this explainability requirement, so option B is correct.