QuestionQ113

VerifiedAI Safety, Security, and Governance

A company is building a generative AI (GenAI) application that uses Amazon Bedrock foundation models (FMs). The solution must meet the following requirements:

  • Authenticate users through Microsoft Entra ID.
  • Ensure that all foundation model (FM) API calls travel only over private network paths.
  • Restrict application access so that each department can use only specific model families.
  • Provide a comprehensive audit trail of all model interactions.

Which solution will meet these requirements?

Explanation

Federating Microsoft Entra ID with IAM through SAML lets users assume IAM roles based on their identity, and department-specific IAM roles can be scoped with policy conditions that permit only certain ModelId values, enforcing model-family restrictions per department. Creating AWS PrivateLink interface VPC endpoints for the Amazon Bedrock runtime service ensures that InvokeModel and related API calls travel entirely over the AWS private network rather than the public internet. For auditing, AWS CloudTrail records management and data events for Bedrock API calls (caller identity, modelId, timestamps), while Amazon Bedrock model invocation logging separately captures the full request and response payloads (prompts and completions) to Amazon S3/CloudWatch Logs — together these provide the comprehensive audit trail of model interactions that AWS documentation recommends for governance of Bedrock usage.

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!