A team trains an Apache MXNet digit classifier in SageMaker and wants (1) an alert if the model starts overfitting and (2) auditors to be able to review SageMaker API activity. What is the simplest solution with minimal code and steps?

The correct solution addresses both requirements efficiently. Enabling AWS CloudTrail to record SageMaker API calls to S3 automatically provides a detailed log for auditors without needing custom code or Lambda functions. For detecting overfitting, the simplest approach is to add code within the training script to emit a custom CloudWatch metric (e.g., monitoring the difference between training and validation accuracy). A CloudWatch alarm can then be configured to trigger an SNS notification when this metric crosses a predefined threshold, indicating overfitting. Incorrect options: Creating a Lambda function to log SageMaker API calls to S3 is unnecessary; CloudTrail handles this natively and more comprehensively. Creating a Lambda function to send SageMaker API call logs to CloudTrail is redundant; CloudTrail already records API calls. Configuring an SNS topic to receive notifications when the model is overfitting is incomplete; it lacks the mechanism to detect overfitting (the custom CloudWatch metric and alarm).