What's changed: Created Professional Cloud Security Engineer Chapter 3 (Domain 3 "Data protection": Sensitive Data Protection (PII discovery/redaction/pseudonymization/FPE), restricting BigQuery/Cloud Storage/Cloud SQL, Secret Manager, instance metadata protection; default encryption/CMEK/Cloud EKM, software/hardware keys (Cloud HSM), key rotation/revocation/import, object lifecycle, Confidential Computing; AI/ML security, IaaS/PaaS training models, Vertex AI security controls (VPC Service Controls/PSC/CMEK)).
3.2Encryption and securing AI workloads
Understand use cases for Google default encryption, customer-managed encryption keys (CMEK), and Cloud External Key Manager (EKM); choosing software vs hardware keys; creating/managing CMEK/EKM keys (rotation/revocation/import); encryption methods per use case; Cloud Storage object lifecycle; Confidential Computing; and security/privacy controls for AI/ML systems and Vertex AI.
Choose encryption by "who manages the key." Increase key sovereignty default→CMEK→EKM and pick by use case. AI workloads are in scope too.
3.2.1Encryption at rest, in transit, and in use
Google Cloud always encrypts data at rest with default encryption. To manage keys yourself, use CMEK (customer-managed keys in Cloud KMS); to hold keys in an external key-management backend for greater sovereignty, use Cloud External Key Manager (Cloud EKM). Choose software keys vs FIPS-compliant tamper-resistant hardware keys (HSM) by requirement, and manage rotation, revocation, and import (BYOK). Protect data in use (in memory) with Confidential Computing (processing in encrypted VMs/nodes). Automate retention/deletion with Cloud Storage object lifecycle. Map "external keys for maximum sovereignty = Cloud EKM" and "protect data in memory = Confidential Computing."
3.2.2Securing AI workloads
Protect AI/ML systems from unintentional exploitation of training data/models (theft/poisoning of data/models). Recognize that security requirements differ between IaaS-hosted and PaaS-hosted training models, and for Vertex AI combine IAM, VPC Service Controls, CMEK, private connectivity (PSC), and data redaction (SDP) to implement security/privacy controls. Map "protect training data/models = SDP + VPC Service Controls + CMEK" and "run Vertex AI inside a perimeter = VPC Service Controls/PSC."
Common: requirement → means. E.g., "manage keys yourself" = CMEK (Cloud KMS); "hold keys externally for max sovereignty" = Cloud EKM; "FIPS-compliant tamper-resistant keys" = Cloud HSM (hardware keys); "encrypt data in use (in memory)" = Confidential Computing; "automate Cloud Storage retention/deletion" = object lifecycle; "keep Vertex AI inside a perimeter" = VPC Service Controls/PSC.
Watch the mix-ups: (1) CMEK (customer-managed in Cloud KMS) vs Cloud EKM (keys held externally) differ in sovereignty. (2) Encryption at rest/in transit vs protection in use (Confidential Computing) are different surfaces. (3) Key revocation effectively makes ciphertext inaccessible—operate with that impact in mind.
3.2.3Section summary
- Key mgmt = default→CMEK (Cloud KMS)→Cloud EKM (external, max sovereignty); software/hardware keys by need
- Protect data in use = Confidential Computing; Cloud Storage object lifecycle automates retention
- AI workloads = SDP + VPC Service Controls + CMEK + PSC; run Vertex AI inside a perimeter
Sign in to track progress — Log in.
Quick check
(just a quick review)Q1. To create/rotate/manage at-rest encryption keys yourself in Cloud KMS, which is best?
Q2. For compliance, to hold encryption keys outside Google (your/third-party key-management backend) for maximum sovereignty, which is best?
Q3. To encrypt data even in use (in memory) and protect it from the host/other tenants, which is best?
Q4. To protect encryption keys in FIPS 140-2 Level 3 tamper-resistant hardware, which is best?
Q5. To protect Vertex AI training data/models against exfiltration and for privacy, which combination is best?