Utilize the Cloud Data Loss Prevention (DLP) API to identify sensitive data, then apply Dataflow with the DLP API to encrypt sensitive values using Format Preserving Encryption.

Before training, filter out columns containing sensitive data in BigQuery and establish an authorized view to restrict access to sensitive information.

Employ Dataflow to extract columns with sensitive data from BigQuery and then randomly shuffle the values within each sensitive column.

Scan for sensitive data using the Cloud Data Loss Prevention (DLP) API, then use Dataflow to replace all sensitive data with AES-256 encryption, including a salt.

Implement both Format Preserving Encryption for columns used in model training and AES-256 encryption for columns not directly used in training but required for compliance.