CCNA Security, Compliance, and Governance for AI Solutions Questions

Macie uses managed data identifiers and pattern matching to scan S3 objects and report PII such as names, addresses and credit card numbers. It satisfies the requirement to detect sensitive data in the training dataset before training begins, without modifying the data.

Grounding check measures whether each response is supported by the retrieved source passages, flagging or filtering ungrounded statements. This directly reduces hallucination by tying outputs to the proprietary knowledge base documents rather than model memory.

Cross-account access requires permissions at both layers: the S3 bucket policy grants the partner account read access to the objects, while the SSE-KMS key policy must separately allow that account to decrypt using the customer managed key. Missing either blocks access.

OpenSearch Serverless supports specifying a customer managed AWS KMS key in the collection's encryption policy at creation. This key encrypts the vector index at rest and allows the security team to manage rotation and revoke access by disabling or deleting the key. This directly satisfies the requirement for customer-controlled encryption of the vector store used by the RAG application.

Topic restrictions define denied subjects, so the guardrail evaluates prompts and responses against them and blocks investment-advice conversations. Content filters target harmful categories such as hate or violence, whereas denied topics are the mechanism for excluding specific subject matter.

The grounding check compares each model response against the retrieved source chunks and flags or blocks claims unsupported by them, directly reducing hallucination. It satisfies the grounding requirement by validating output against provided documents rather than relying on the model's parametric knowledge.

SageMaker Model Monitor runs scheduled bias and drift jobs against a deployed endpoint, integrating Clarify's bias metrics into automated monitoring schedules. This satisfies the stem's requirement for automatic, recurring bias detection, whereas Clarify alone provides analysis without native scheduling.

Disparate impact is a measured bias metric, so the remedy is mitigation, not monitoring. Reweighing training data or applying bias mitigation algorithms directly addresses the detected disparity in the protected group before redeployment, satisfying the requirement to correct the bias rather than merely report it.

Denied topics let you define a set of undesirable subjects, described in natural language, that Bedrock Guardrails blocks the model from discussing. Configuring a denied topic covering competitors prevents the assistant generating responses about them, meeting the restriction requirement.

S3 VPC endpoints keep S3 traffic on the AWS private network within the Region, and the private subnet without internet access prevents any egress outside it. This satisfies the data residency constraint while avoiding NAT gateway data transfer costs.

Placing the notebook in a VPC lets security groups act as stateful IP filters, permitting only the corporate CIDR ranges. Direct internet access is removed, so the network-origin constraint is enforced at the elastic network interface.

Amazon Macie uses machine learning and pattern matching to automatically discover, classify and alert on sensitive data such as PII within S3 buckets, directly satisfying the requirement to scan training datasets before use. It continuously evaluates bucket contents against managed and custom data identifiers, providing the visibility needed prior to model training.

Guardrails evaluate prompts and responses against configured policies, and the sensitive information filters detect and block entities such as names, addresses, and other PII patterns. Because Guardrails act inline during InvokeModel and Converse calls, they stop disallowed content before it reaches the model or returns to the user, matching the real-time blocking requirement.

Model evaluation jobs provide systematic, repeatable measurements before launch, letting the team compare models or configurations on quality and safety metrics. The results create evidence for a go or no-go decision and a baseline for later comparisons. This is a pre-deployment governance control that complements runtime filtering rather than replacing it.

Amazon Bedrock Guardrails allows you to define policies that filter and mask sensitive information in model inputs and outputs. You can configure sensitive information filters to detect and redact PII such as account numbers and social security numbers. This directly meets the compliance requirement by preventing PII from appearing in responses and automatically masking it in real time.

Data capture on a SageMaker endpoint records real-time inference requests and responses to S3, producing the live production dataset that drift detection requires. Without captured data, Model Monitor has nothing to compare against the training baseline.

Topic restrictions define denied topics using natural-language descriptions and example phrases, blocking discussion of named subjects such as proprietary internal projects. Content filters address harmful categories, not specific business topics, so topic restrictions satisfy this constraint.

Reviewing the third-party provider's privacy and handling documentation within the AWS Bedrock service description reveals whether prompts, completions, or fine-tuning data are retained, logged, or used for model training. This directly addresses the stem's constraint: understanding the provider's data handling practices and whether they can access company data.

SageMaker Clarify runs bias metrics such as disparate impact against deployed endpoints and emits them to CloudWatch, detecting bias after deployment. This satisfies the post-deployment monitoring requirement, which pre-training Clarify analysis alone would not cover.

Amazon SageMaker Model Registry provides a centralised catalogue for versioning ML models, tracking approval status, and managing their lifecycle through to decommissioning. It satisfies the stem's governance requirements by recording model metadata and lineage, while drift monitoring is handled by SageMaker Model Monitor alongside it.

Amazon Macie uses machine learning and pattern matching to automatically discover, classify and alert on sensitive data such as credit card numbers stored in Amazon S3. It directly satisfies the requirement to scan S3 training datasets before use, providing continuous visibility without manual inspection or custom scripts.

Lake Formation provides table- and column-level permission grants over S3 data registered in its Data Catalog, meeting the fine-grained access requirement. IAM policies alone cannot restrict individual columns, so Lake Formation is the only service here that enforces column-level control across business units.

Bedrock model invocation logging captures full request and response payloads for each foundation model call, and directing output to S3 with a one-year lifecycle retention policy satisfies the compliance requirement for auditable, retained inputs and outputs.

Concept drift occurs when the statistical relationship between input features and the target variable changes, degrading accuracy even if input distributions remain stable. Retraining with recent data captures the new feature-to-label mapping, directly addressing the declining accuracy described in the stem.

Specifying a KMS customer-managed key for the training job and encrypting the source S3 bucket with the same key satisfies the policy that all data at rest uses a customer-managed key. SageMaker's default encryption uses AWS-owned keys, which the policy excludes.

Column-level permissions in Lake Formation grant or deny access to individual columns within a table, so users querying sensitive ML training data see only authorised fields. This satisfies the requirement to restrict access to specific columns rather than the whole table.

Model Registry versions approved model artifacts with approval status, while ML Lineage Tracking automatically captures entities and associations such as training datasets, processing jobs, trial components, and endpoints. Together they let the company connect a deployed endpoint back to the exact artifact, image, and data version, which is precisely the traceability an auditor requests.

Contextual grounding check evaluates responses against the supplied source documents and a query, scoring grounding and relevance. It blocks or flags outputs containing unsupported claims, directly enforcing the requirement that the model use only provided source material rather than generating fabricated content.

Guardrails applies configurable content filters and denied-topic policies directly to Bedrock model invocations, blocking harmful or biased output before it reaches the application. This satisfies the requirement to enforce content safety policies at inference time, which prompt engineering alone cannot guarantee.

Data retention is governed by each third-party provider's own terms, not by Amazon Bedrock itself. Reviewing the provider's data handling policy and selecting a model whose terms guarantee no retention directly addresses the privacy concern.

Amazon Bedrock Guardrails allows you to define denied topics to block discussions on specific subjects like competitors or medical advice. It also provides content filters to block harmful content. Guardrails can be configured to log blocked prompts to CloudWatch or S3, enabling monitoring and analysis. This directly meets the requirements.

SageMaker Pipelines orchestrates the end-to-end ML workflow as code, automating retraining, evaluation and conditional deployment steps. Within a governance framework it enforces repeatable, auditable lifecycle transitions, complementing Model Registry versioning and Model Monitor drift detection rather than duplicating them.

SageMaker Clarify provides tools to detect bias and explain model predictions. It generates feature attribution explanations using SHAP values, which show how each input feature contributes to the model's output. This directly meets the requirement for explainability and documentation of model decisions, helping the company comply with regulations that demand transparency.

SageMaker Clarify provides feature attribution explanations using SHAP values, which show the contribution of each feature to a model's prediction. This allows the company to explain individual credit decisions to customers and auditors. Clarify can be integrated with SageMaker endpoints to generate explanations in real time, meeting the compliance requirement.

CloudTrail records API activity as management and data events, capturing every Bedrock API call with caller identity, timestamp and source IP. Enabling a trail delivers the audit logging the firm requires, satisfying the compliance constraint without altering application code.

Amazon Bedrock Guardrails provides configurable content filters to block hate speech, profanity, and other harmful content. It also includes prompt attack detection to identify jailbreak attempts. When model invocation logging is enabled, Bedrock logs the full request and response, including blocked interactions, allowing for review. This native integration meets all requirements without custom code.

Amazon Macie uses machine learning and pattern matching to automatically discover, classify and report sensitive data such as PII in S3. It continuously evaluates buckets and raises findings, directly satisfying the requirement to detect and report PII before SageMaker training begins.

Sensitive information filters detect and redact PII such as names, addresses and credit card numbers in prompts, directly satisfying the requirement to remove personal data before it reaches the foundation model. Configuring these filters with appropriate PII entities and action set to Mask achieves the redaction.

AWS CloudTrail records Bedrock control-plane and data-plane API activity, including model invocations and guardrail evaluations, as audit events. It satisfies the compliance requirement to capture every API call by logging requests, identities, timestamps and source IPs to an immutable trail, which can then be queried or exported for auditing.

Sensitive information filters detect and redact personally identifiable information, matching the stem's requirement to block social security numbers in generated summaries. Bedrock Guardrails applies these filters to both prompts and responses, so PII is caught before reaching users. Unlike content filters, which target harmful categories such as hate or violence, this mechanism specifically addresses data privacy.

Amazon SageMaker Clarify generates post-deployment bias reports by monitoring live endpoint traffic and computing metrics such as disparate impact against baseline data. This satisfies the stem's requirement to detect bias in predictions on new data after deployment, which pre-training Clarify analysis alone cannot cover.

Enabling model invocation logging in Bedrock and designating an S3 bucket (with optional CloudWatch Logs) captures both the request input and the model output for every API call, directly meeting the compliance audit requirement to record full invocation content.

Amazon CloudWatch Logs is a supported destination for Bedrock model invocation logging, satisfying the audit and troubleshooting requirement. It captures text, image, and embedding invocation data as log events, enabling near-real-time querying and retention policies for compliance reviews.

A cross-account role lets the vendor's own principals call AssumeRole and receive temporary credentials, so no long-term secrets are shared. The external ID condition guards against the confused deputy problem, and because the vendor assumes a distinct role, CloudTrail records its activity separately. This satisfies both the no-shared-credentials policy and the auditability requirement.

AWS KMS customer managed keys give the company full control over key creation, rotation, and access policies. They can enable automatic rotation and define granular permissions. This meets the requirement for a company-managed key that can be rotated independently, and it integrates with S3 server-side encryption to protect data at rest.

Network isolation disables outbound network access from the training container, preventing data exfiltration over the internet. VPC interface endpoints keep calls to services such as Amazon S3 and SageMaker on private IP addresses inside the VPC. Together they satisfy both halves of the policy without exposing data or control traffic to public routes.

Amazon Bedrock Knowledge Bases supports metadata filtering, allowing you to attach metadata to documents and filter retrieval based on user attributes. By passing user department as a filter, the assistant only retrieves authorized documents. This provides fine-grained access control at the document level.

Bedrock's model invocation logging natively supports dual destinations: you enable it once and specify both an S3 bucket for durable audit storage and a CloudWatch Logs group for real-time monitoring. This satisfies the stem's requirement for centralised S3 retention plus live CloudWatch visibility, without custom pipelines or duplicate instrumentation.

SageMaker Clarify detects bias in deployed models through bias metrics on predictions and features, and integrates with Model Monitor for ongoing post-deployment checks. This satisfies the requirement to monitor bias over time rather than only during training.

Bedrock resource-based policies attach directly to a specific base model, letting you define which principals may invoke it and under what conditions. This satisfies the stem's constraint of restricting IAM roles per model, since identity-based policies alone cannot scope permissions to an individual Bedrock model resource.

CloudWatch alarms on accuracy metrics detect drift and degradation during live inference, satisfying the governance requirement to monitor deployed models continuously. This provides the observability needed to trigger retraining or rollback decisions across the model lifecycle.

A VPC interface endpoint (AWS PrivateLink) provisions an elastic network interface with a private IP inside the subnet, letting inference calls reach SageMaker without traversing the public internet. Disabling public access enforces the stem's constraint that endpoints remain reachable only from the private subnet, satisfying the no-internet requirement.

AWS KMS provides the customer-managed keys that encrypt SageMaker training data in S3 and model artefacts at rest, satisfying the at-rest encryption requirement. KMS integrates natively with SageMaker and S3, unlike services handling in-transit encryption or access control.

Word filters block or mask prompts and responses containing configured custom terms, so adding internal project names stops the model emitting them. This is distinct from managed word filters, which target profanity, and from denied topics, which cover broader subject areas.

The bucket policy is the resource-based control that explicitly overrides the current prohibition on external accounts. Granting the cross-account SageMaker execution role s3:GetObject and s3:ListBucket authorises that principal directly, satisfying the cross-account access requirement while keeping the bucket closed to everyone else.

AWS Glue crawlers scan the underlying data, infer schemas and register tables in the Data Catalog. Lake Formation permissions are granted against those catalogued tables and columns, so cataloguing is the prerequisite step enabling column-level access control.