🗺️ Roadmap: The AI Infra Security & Compliance¶
Below is the syllabus timeline of lessons in this module. Track your study progress here.
📈 Progress¶
📁 35 Securing the AI Infrastructure Stack¶
- 35.1a Model theft: protecting valuable trained model weights from unauthorized access
- 35.1b Data poisoning: adversarial manipulation of training datasets
- 35.1c Supply chain attacks: compromised containers and malicious packages in NGC
- 35.1d Cryptojacking: unauthorized GPU compute usage for cryptocurrency mining
- 35.2a Verifying container image signatures: Docker Content Trust (DCT) and Notary
- 35.2b SBOM (Software Bill of Materials) for AI containers
- 35.2c CVE scanning: Trivy, Clair, and Grype for container vulnerability scanning
- 35.2d Runtime security: Falco for detecting anomalous GPU process behavior
- 35.3a RBAC (Role-Based Access Control): controlling who can submit GPU jobs
- 35.3b Pod Security Standards: restricted, baseline, and privileged policies
- 35.3c Network policies: restricting east-west traffic between AI namespaces
- 35.3d Secrets management: HashiCorp Vault and sealed secrets for API keys and model credentials
- 35.4a SOC 2 Type II: security, availability, and confidentiality controls
- 35.4b HIPAA: healthcare AI deployments and PHI in training data
- 35.4c FedRAMP: U.S. government AI infrastructure authorization
- 35.4d Export controls (EAR/ITAR): implications for H100 and A100 deployment in restricted countries
- 35.4e EU AI Act: compliance considerations for AI system operators