Ethics & Safety · Advanced
Securing AI Workloads: IAM, VPC Service Controls, Model Armor and Data Governance
The controls that get a generative AI workload past a security review: least-privilege IAM, a VPC Service Controls perimeter, CMEK, Model Armor against prompt injection, and honest answers about data use.
- 1.4 hours lesson
- advanced level
- Google Cloud AI Track
- Narrated presentation
What you’ll learn
- Least-privilege IAM with roles/aiplatform.user and per-workload service accounts
- VPC Service Controls perimeters to prevent data exfiltration
- Private Service Connect / Private Google Access and CMEK
- Google Cloud generative AI data governance: prompts not used to train foundation models
- Model Armor templates for prompt injection, jailbreak, sensitive data and malicious URLs
- Indirect prompt injection and architectural blast-radius limits
- Regional data residency and enabling Data Access audit logs
Inside this lesson
A narrated, slide-by-slide presentation with 1 prediction exercise, 1 knowledge check, 3 videos. You answer as you go, and the platform checks your work.