BrightRetail's MLOps team wants data scientists in SageMaker Studio to be able to launch training jobs against a specific S3 training prefix, but the team must prevent those users from listing or reading other S3 prefixes in the same bucket and from changing KMS key policies. Which least-privilege IAM design will accomplish this?

This option correctly applies the principle of least privilege. By creating a dedicated execution role with granular S3 read access to only the required prefix and KMS decrypt permissions, the training jobs have only the necessary permissions. Allowing Studio users sagemaker:CreateTrainingJob and iam:PassRole for only this specific execution role enables them to launch jobs with the restricted permissions without directly accessing S3 or KMS. The other options are incorrect because: Attaching a broad S3 read policy to the Studio user profile role violates least privilege by allowing access to other prefixes. Giving the Studio user profile role AdminAccess is a severe security risk and relies on unenforceable organizational policy. Granting sts:AssumeRole without iam:PassRole prevents users from passing that role to a SageMaker training job, making it impossible to launch the job with those specific permissions.