An ML specialist is running distributed training of a BERT model on eight SageMaker instances. The training data resides in S3. What combination of steps should be taken to protect data during distributed training? (Choose three.)

Running distributed training inside a private VPC with encrypted inter-container traffic ensures network isolation and protects data in transit between instances. Creating an S3 VPC endpoint with proper configurations allows SageMaker instances to access S3 data privately without traversing the public internet, enhancing security. Assigning an IAM role with read-only access to S3 for the SageMaker jobs adheres to the principle of least privilege, preventing unauthorized data modification or deletion. Running across multiple VPCs and peering is unnecessarily complex for a single training job. A NAT gateway is for outbound internet access from private subnets, not direct S3 access or inter-container communication. Adding an inbound security group rule is too broad and less secure than VPC-level controls and encryption.