A media streaming company is using Amazon Redshift ML in its primary AWS account to build recommendation models for personalized content suggestions. The training data for these models, which includes user watch history and preferences, is stored in an Amazon S3 bucket in a secondary AWS account. The company requires a secure pipeline that allows Redshift ML in the primary account to access this training data while ensuring that no public IPv4 addresses are used in the data transfer process. What do you recommend?