Back to Checks
aws

Amazon SageMaker training job has network isolation enabled

sagemaker_training_jobs_network_isolation_enabled
Service: sagemakerSeverity: high
OfficialBuilt-inVerified
Categories:trust-boundariesgen-aiinternet-exposed
Resource Type:Other

SageMaker training jobs have network isolation enabled, preventing the training container from making any inbound or outbound network calls during execution

Prowler logoProwlerprowler-provider-aws

Risk

Without network isolation, training code can reach the internet or internal services, enabling:

  • Data exfiltration of datasets and model artifacts
  • Supply-chain compromise via untrusted downloads
  • C2 beacons and resource abuse This harms confidentiality and integrity, and may impact availability.

Run this check with Prowler

Run in Prowler Cloud
Run
prowler aws --checks sagemaker_training_jobs_network_isolation_enabled

Recommendation

Enable network isolation for training jobs by default. If network access is required, enforce least privilege and defense in depth:

  • Use private networking and VPC endpoints
  • Restrict egress and narrow IAM permissions
  • Prepackage dependencies to avoid external downloads

Remediation

Terraform

# SageMaker Training Job with network isolation enabled
resource "aws_sagemaker_training_job" "<example_resource_name>" {
  name     = "<example_resource_name>"
  role_arn = "<example_resource_id>"

  algorithm_specification {
    training_image      = "<example_resource_id>"
    training_input_mode = "File"
  }

  output_data_config {
    s3_output_path = "s3://<example_resource_name>/"
  }

  resource_config {
    instance_type  = "ml.m5.large"
    instance_count = 1
    volume_size_gb = 10
  }

  stopping_condition {
    max_runtime_in_seconds = 3600
  }

  input_data_config {
    channel_name = "training"
    data_source {
      s3_data_source {
        s3_data_type = "S3Prefix"
        s3_uri       = "s3://<example_resource_name>/input/"
      }
    }
  }

  enable_network_isolation = true # Critical: blocks all network access from the training container
}

NativeIaC

# CloudFormation: SageMaker Training Job with network isolation enabled
Resources:
  <example_resource_name>:
    Type: AWS::SageMaker::TrainingJob
    Properties:
      TrainingJobName: <example_resource_name>
      RoleArn: <example_resource_id>
      AlgorithmSpecification:
        TrainingImage: <example_resource_id>
        TrainingInputMode: File
      OutputDataConfig:
        S3OutputPath: s3://<example_resource_name>/
      ResourceConfig:
        InstanceType: ml.m5.large
        InstanceCount: 1
        VolumeSizeInGB: 10
      StoppingCondition:
        MaxRuntimeInSeconds: 3600
      InputDataConfig:
        - ChannelName: training
          DataSource:
            S3DataSource:
              S3DataType: S3Prefix
              S3Uri: s3://<example_resource_name>/input/
      EnableNetworkIsolation: true  # Critical: blocks all network access from the training container

Other

1In the AWS Console, open SageMaker > Training jobs
2Click Create training job
3Fill required fields (role, image, input/output, resources)
4Enable the setting: Enable network isolation
5Create the job
6Note: You cannot edit an existing training job to enable this; create a new job with network isolation and retire non-compliant jobs.

Check Code

from prowler.lib.check.models import Check, Check_Report_AWS
from prowler.providers.aws.services.sagemaker.sagemaker_client import sagemaker_client


class sagemaker_training_jobs_network_isolation_enabled(Check):
    def execute(self):
        findings = []
        for training_job in sagemaker_client.sagemaker_training_jobs:
            report = Check_Report_AWS(metadata=self.metadata(), resource=training_job)
            report.status = "PASS"
            report.status_extended = f"Sagemaker training job {training_job.name} has network isolation enabled."
            if not training_job.network_isolation:
                report.status = "FAIL"
                report.status_extended = f"Sagemaker training job {training_job.name} has network isolation disabled."

            findings.append(report)

        return findings

References