A medical organization stores clinical records containing PII and PHI. An ML engineer must ensure that neither PII nor PHI is used for model training. Which approach satisfies this requirement?

The correct approach is to use Amazon Comprehend for PII and Amazon Comprehend Medical for PHI. Amazon Comprehend is a natural language processing (NLP) service that can identify and redact PII from unstructured text. Similarly, Amazon Comprehend Medical is specifically designed to detect and redact PHI from clinical notes and other medical text. These services are purpose-built for this task, ensuring accurate and compliant data masking before model training. Storing data in S3 and using AWS Glue DataBrew to mask PII and PHI is a viable option, but it requires more manual configuration and rule definition compared to the specialized Comprehend services. Loading data into Amazon Redshift and using SQL stored procedures for classification and masking is less suitable for unstructured text data like clinical records, as SQL is not optimized for NLP tasks. Creating a Lambda function to encrypt PII and PHI is insufficient because encryption only protects data at rest; the data would still need to be decrypted for training, exposing the sensitive information. The requirement is to prevent the use of PII/PHI for training, not just to encrypt it.