Security

This guide covers security best practices for the MLOps platform.

Overview

Security considerations:

Access Control: Who can access what?
Network Security: Protecting communication
Data Security: Protecting sensitive data
Secrets Management: Handling credentials

AWS IAM

Principle of Least Privilege

The pipeline IAM role has minimal permissions:

# S3 access - only to specific bucket
resource "aws_iam_role_policy" "mlops_s3_access" {
  policy = jsonencode({
    Statement = [{
      Effect = "Allow"
      Action = [
        "s3:GetObject",
        "s3:PutObject",
        "s3:DeleteObject",
        "s3:ListBucket"
      ]
      Resource = [
        "arn:aws:s3:::064592191516-mlflow",
        "arn:aws:s3:::064592191516-mlflow/*"
      ]
    }]
  })
}

IAM Best Practices

Use roles, not users for EC2 instances
Enable MFA for console access
Rotate credentials regularly
Audit with CloudTrail

Network Security

Security Groups

Restrict inbound traffic:

resource "aws_security_group" "mlops_pipeline" {
  # SSH - restrict to your IP
  ingress {
    from_port   = 22
    to_port     = 22
    protocol    = "tcp"
    cidr_blocks = ["YOUR_IP/32"]
  }

  # MLflow - internal only or VPN
  ingress {
    from_port   = 5000
    to_port     = 5000
    protocol    = "tcp"
    cidr_blocks = ["10.0.0.0/8"]  # VPC only
  }

  # Outbound - allow all
  egress {
    from_port   = 0
    to_port     = 0
    protocol    = "-1"
    cidr_blocks = ["0.0.0.0/0"]
  }
}

VPN Access

For production, use VPN or bastion host:

                    Internet
                        │
                        ▼
            ┌───────────────────┐
            │   Bastion Host    │
            │   (Public Subnet) │
            └─────────┬─────────┘
                      │ SSH
                      ▼
            ┌───────────────────┐
            │   Pipeline EC2    │
            │  (Private Subnet) │
            └───────────────────┘

SSH Key Management

# Generate strong key
ssh-keygen -t ed25519 -C "mlops-pipeline"

# Set proper permissions
chmod 600 ~/.ssh/mlops-pipeline-key

# Use SSH agent
eval $(ssh-agent -s)
ssh-add ~/.ssh/mlops-pipeline-key

Data Security

S3 Bucket Security

# Enable versioning
resource "aws_s3_bucket_versioning" "mlflow" {
  bucket = aws_s3_bucket.mlflow_artifacts.id
  versioning_configuration {
    status = "Enabled"
  }
}

# Enable encryption
resource "aws_s3_bucket_server_side_encryption_configuration" "mlflow" {
  bucket = aws_s3_bucket.mlflow_artifacts.id

  rule {
    apply_server_side_encryption_by_default {
      sse_algorithm = "AES256"
    }
  }
}

# Block public access
resource "aws_s3_bucket_public_access_block" "mlflow" {
  bucket = aws_s3_bucket.mlflow_artifacts.id

  block_public_acls       = true
  block_public_policy     = true
  ignore_public_acls      = true
  restrict_public_buckets = true
}

EBS Encryption

All EBS volumes are encrypted:

block_device_mappings {
  device_name = "/dev/sda1"
  ebs {
    volume_size           = 100
    volume_type           = "gp3"
    encrypted             = true
    delete_on_termination = true
  }
}

Data at Rest

S3: Server-side encryption (AES-256)
EBS: Encrypted volumes
MLflow DB: SQLite with encrypted volume

Data in Transit

HTTPS for all API calls
SSH for remote access
TLS for MLflow (when exposed)

Secrets Management

AWS Secrets Manager

Store sensitive credentials:

# Create secret
aws secretsmanager create-secret \
    --name mlops/database-password \
    --secret-string "your-password"

# Retrieve in code
import boto3

def get_secret(secret_name):
    client = boto3.client('secretsmanager')
    response = client.get_secret_value(SecretId=secret_name)
    return response['SecretString']

Environment Variables

Don’t hardcode secrets:

# Bad
API_KEY = "sk-1234567890"

# Good
import os
API_KEY = os.environ.get("API_KEY")

.gitignore

Ensure secrets aren’t committed:

# Secrets
.env
*.pem
credentials.json
secrets/

Audit and Compliance

CloudTrail

Enable CloudTrail for audit logs:

aws cloudtrail create-trail \
    --name mlops-audit \
    --s3-bucket-name 064592191516-audit-logs

Access Logging

Enable S3 access logging:

resource "aws_s3_bucket_logging" "mlflow" {
  bucket = aws_s3_bucket.mlflow_artifacts.id

  target_bucket = aws_s3_bucket.logs.id
  target_prefix = "s3-access-logs/"
}

Security Scanning

Scan for vulnerabilities:

# Scan Python dependencies
pip install safety
safety check -r requirements.txt

# Scan Docker images
trivy image mlflow-sklearn:latest

# Scan Terraform
tfsec deploy/aws/064592191516/us-east-1/01-infrastructure/

Checklist

Before Deployment

Review IAM permissions
Restrict security group rules
Enable encryption (S3, EBS)
Configure CloudTrail
Set up VPN/bastion if public

Regular Maintenance

Rotate SSH keys quarterly
Review CloudTrail logs weekly
Update dependencies monthly
Security scan before releases
Audit IAM permissions quarterly

Incident Response

Detect: Monitor CloudWatch alerts
Contain: Isolate affected resources
Investigate: Review CloudTrail logs
Remediate: Fix vulnerability
Document: Update runbooks