Cloud ML Platform 安全 (AWS/Azure/GCP)

安全 comparison of cloud ML platforms including AWS SageMaker, Azure Machine Learning, and Google Vertex AI. IAM configuration, data security, model serving, and platform-specific attack surfaces.

雲端 ML platforms (AWS SageMaker, Azure Machine Learning, Google Vertex AI) provide managed infrastructure for 訓練, deploying, and serving ML models — including LLMs. Each platform inherits its parent 雲端's 安全 model while adding ML-specific 攻擊面. Red teaming 雲端 ML deployments requires 理解 both 雲端 安全 fundamentals and ML-specific 漏洞.

Platform Comparison 概覽

IAM Misconfiguration Risks

IAM misconfigurations are the most common and highest-impact 漏洞 class in 雲端 ML deployments.

AWS SageMaker IAM Risks

# DANGEROUS: Overly permissive SageMaker execution role
overly_permissive_policy = {
    "Version": "2012-10-17",
    "Statement": [{
        "Effect": "Allow",
        "Action": "sagemaker:*",
        "Resource": "*"
    }, {
        "Effect": "Allow",
        "Action": "s3:*",      # Full S3 access — can read ALL buckets
        "Resource": "*"
    }, {
        "Effect": "Allow",
        "Action": "iam:PassRole",  # Can pass any role
        "Resource": "*"
    }]
}
 
# This role can:
# - Access any S3 bucket (訓練資料, other projects' data)
# - Create endpoints with any role attached
# - Modify any SageMaker resource
# - Potentially escalate privileges via PassRole

Azure ML IAM Risks

Google Vertex AI IAM Risks

Data 安全 Across the ML Lifecycle

Data Collection → Preparation → Training → Model Storage → Serving → Inference
      ↑               ↑            ↑            ↑              ↑          ↑
  Data source     Feature       Training    Model         Endpoint    Response
  access control  store access  data in     artifact      auth        data
  認證  encryption    compute     encryption    rate limit  leakage

Training Data 安全

Model Artifact 安全

Model Serving 安全

Endpoint Authentication

Endpoint 攻擊 Surface

1. 1

   測試 Authentication Bypass

   Attempt to invoke model endpoints without valid credentials. Check for misconfigured API gateways, missing auth on health check endpoints, and default credentials.

2. 2

   測試 輸入 Validation

   Send malformed, oversized, and 對抗性 inputs to model endpoints. Check for error messages that reveal model architecture, framework versions, or infrastructure details.

3. 3

   測試 for Model Extraction

   Systematically query the endpoint to extract model behavior. With enough queries, 攻擊者 can build a functional copy of 模型 (model stealing). Check rate limiting and query logging.

4. 4

   測試 for Data Leakage in Responses

   Check whether model responses contain 訓練資料, PII, or other sensitive information. 測試 for membership 推論 (determining whether a specific data point was in the 訓練 set).

5. 5

   測試 Network Isolation

   Verify that model endpoints cannot be reached from unauthorized networks. Check VPC configurations, 安全 groups, and network policies.

Notebook Environment 安全

ML notebook environments (SageMaker Studio, Azure ML Compute, Vertex AI Workbench) are high-value targets 因為 they typically have broad data access and code execution capabilities:

Cross-Platform 安全 Checklist

For related topics, see Infrastructure 安全, Supply Chain 安全, and LangChain & LlamaIndex 安全.

相關主題

• Infrastructure 安全: API 安全 -- API-level 安全 測試 for ML endpoints
• Supply Chain 安全 -- model artifact and dependency integrity
• LangChain & LlamaIndex 安全 -- framework-specific 漏洞 on 雲端 platforms
• API Provider 安全 Comparison -- comparing 安全 controls across LLM API providers

參考文獻

• "AWS SageMaker 安全 最佳實務" - Amazon Web Services (2024) - AWS guidance on securing ML workloads including IAM, networking, and encryption
• "Azure Machine Learning 安全 Baseline" - Microsoft (2024) - 安全 controls and configuration for Azure ML deployments
• "Google 雲端 Vertex AI 安全 概覽" - Google 雲端 (2024) - 安全 architecture and best practices for Vertex AI workloads
• "CIS Benchmarks for 雲端 ML Services" - Center for Internet 安全 (2024) - Hardening benchmarks for 雲端 ML platform configurations