Question

A financial services company runs a generative AI application on Amazon Bedrock. The security team requires that a specific application role be allowed to invoke only one approved foundation model, and be explicitly prevented from calling any other model or from creating fine-tuning jobs. Which approach best enforces this requirement?

Accepted Answer

Attach an IAM policy to the role that allows bedrock:InvokeModel only on the specific model's Amazon Resource Name (ARN) and grants no other Bedrock permissions.

Answer

Grant the role the AmazonBedrockFullAccess managed policy and rely on Amazon Bedrock Guardrails to block unwanted models.

Answer

Enable AWS CloudTrail logging for all Bedrock API calls so unauthorized model invocations can be reviewed and reversed later.

Answer

Encrypt the model inference data with an AWS KMS key so only authorized users can decrypt the responses.