A financial enterprise is deploying proprietary machine learning models for online inference on Vertex AI. The security team requires a multi-layered security architecture that accomplishes the following: Prevents unauthorized access and data exfiltration of training datasets, model artifacts, and inference requests across network boundaries. Ensures encryption at rest using customer-controlled keys with strict separation of duties between security administrators and AI developers. Enforces least privilege for model execution and provides visibility into all model read and prediction operations. Which architecture should you implement to satisfy these security requirements?

Place the Vertex AI and data storage projects within a common VPC Service Controls perimeter, encrypt model artifacts with CMEK managed in a dedicated Cloud KMS project granting the CryptoKey Encrypter/Decrypter role to service agents, use custom service accounts with least privilege IAM roles for endpoints, and enable Data Access audit logs.: What This Solution Is This architecture represents Google Cloud's recommended end-to-end security design for enterprise AI workloads. It combines perimeter security via VPC Service Controls, cryptographic governance using Customer-Managed Encryption Keys (CMEK), granular identity governance using Identity and Access Management (IAM) with custom service accounts, and operational visibility through Cloud Audit Logs. How It Meets the Requirements Data Exfiltration Prevention: Placing the Vertex AI project, training data projects, and model repository within a common unified VPC Service Controls perimeter ensures that sensitive model artifacts and datasets cannot be exported to unauthorized projects or accessed over public internet routes. Cryptographic Control and Separation of Duties: Hosting Cloud KMS keys in a dedicated Cloud KMS project isolates key lifecycle management to security administrators. Granting the roles/cloudkms.cryptoKeyEncrypterDecrypter role specifically to the Vertex AI and Cloud Storage service agents ensures automated service-level cryptographic operations without granting developers direct key manipulation privileges. Least Privilege and Visibility: Deploying endpoints with custom service accounts restricts endpoint permissions strictly to required dependencies (e.g., specific BigQuery or Cloud Storage paths) rather than using broad default service agents. Enabling Data Access audit logs ensures full auditability of read and prediction operations against models and datasets. Key Advantages and Capabilities Perimeter Defense: VPC Service Controls mitigates data exfiltration risks caused by compromised credentials or malicious insider actions. Granular Identity Isolation: Custom service accounts isolate less trusted code in custom prediction containers from sensitive corporate datasets. Comprehensive Auditing: Data Access audit logging records every API call and data access event for forensic auditing and compliance reporting. Key Lifecycle Control: Centralized Cloud KMS allows administrators to rotate, disable, or cryptographically erase access to model artifacts instantly. Why This Is Optimal This solution implements deep defense across network, identity, cryptographic, and operational logging layers. It strictly adheres to enterprise separation of duties and standard Google Cloud reference architectures for sensitive AI deployments.

StudyCerts

Professional Cloud Architect

Intrigued by the art of cloud architecture? Discover how to design, develop, and manage robust, secure, scalable, and dynamic solutions on Google Cloud as you prepare for the Professional Cloud Architect exam!

Intermediate

Prepare and test your skills

Track Readiness

Set Exam Goal

Intermediate

Prepare and test your skills

Track Readiness

Set Exam Goal

Secure Ai Model Deployment Vpcsc Cmek Iam

Section: Securing AI (e.g., Model Armor, Sensitive Data Protection, and secure model deployment)

A financial enterprise is deploying proprietary machine learning models for online inference on Vertex AI. The security team requires a multi-layered security architecture that accomplishes the following:

Prevents unauthorized access and data exfiltration of training datasets, model artifacts, and inference requests across network boundaries.
Ensures encryption at rest using customer-controlled keys with strict separation of duties between security administrators and AI developers.
Enforces least privilege for model execution and provides visibility into all model read and prediction operations.

Which architecture should you implement to satisfy these security requirements?

Place the Vertex AI and data storage projects within a common VPC Service Controls perimeter, encrypt model artifacts with CMEK managed in a dedicated Cloud KMS project granting the CryptoKey Encrypter/Decrypter role to service agents, use custom service accounts with least privilege IAM roles for endpoints, and enable Data Access audit logs.

Configure IAM condition bindings restricted by corporate IP ranges on Vertex AI endpoints, configure Customer-Supplied Encryption Keys (CSEK) on the Vertex AI model registry, and deploy standard VPC firewall ingress rules to block unauthorized inference traffic.

Deploy Vertex AI endpoints in a shared development project, attach Google Cloud Armor security policies directly to the Vertex AI prediction endpoints, protect data with Google default encryption, and configure Cloud Monitoring metric alerts for prediction volume.

Deploy Identity-Aware Proxy (IAP) in front of public Vertex AI endpoints, perform client-side symmetric encryption on all model artifacts prior to uploading them to Cloud Storage, and route all container egress traffic through Cloud NAT.

Professional Cloud Architect

Intermediate

Prepare and test your skills

Track Readiness

Set Exam Goal

Intermediate

Prepare and test your skills

Track Readiness

Set Exam Goal

Secure Ai Model Deployment Vpcsc Cmek Iam

Section: Securing AI (e.g., Model Armor, Sensitive Data Protection, and secure model deployment)

Prevents unauthorized access and data exfiltration of training datasets, model artifacts, and inference requests across network boundaries.
Ensures encryption at rest using customer-controlled keys with strict separation of duties between security administrators and AI developers.
Enforces least privilege for model execution and provides visibility into all model read and prediction operations.

Which architecture should you implement to satisfy these security requirements?

Explanations

What This Solution Is

This architecture represents Google Cloud's recommended end-to-end security design for enterprise AI workloads. It combines perimeter security via VPC Service Controls, cryptographic governance using Customer-Managed Encryption Keys (CMEK), granular identity governance using Identity and Access Management (IAM) with custom service accounts, and operational visibility through Cloud Audit Logs.

How It Meets the Requirements

Data Exfiltration Prevention: Placing the Vertex AI project, training data projects, and model repository within a common unified VPC Service Controls perimeter ensures that sensitive model artifacts and datasets cannot be exported to unauthorized projects or accessed over public internet routes.
Cryptographic Control and Separation of Duties: Hosting Cloud KMS keys in a dedicated Cloud KMS project isolates key lifecycle management to security administrators. Granting the roles/cloudkms.cryptoKeyEncrypterDecrypter role specifically to the Vertex AI and Cloud Storage service agents ensures automated service-level cryptographic operations without granting developers direct key manipulation privileges.
Least Privilege and Visibility: Deploying endpoints with custom service accounts restricts endpoint permissions strictly to required dependencies (e.g., specific BigQuery or Cloud Storage paths) rather than using broad default service agents. Enabling Data Access audit logs ensures full auditability of read and prediction operations against models and datasets.

Key Advantages and Capabilities

Perimeter Defense: VPC Service Controls mitigates data exfiltration risks caused by compromised credentials or malicious insider actions.
Granular Identity Isolation: Custom service accounts isolate less trusted code in custom prediction containers from sensitive corporate datasets.
Comprehensive Auditing: Data Access audit logging records every API call and data access event for forensic auditing and compliance reporting.
Key Lifecycle Control: Centralized Cloud KMS allows administrators to rotate, disable, or cryptographically erase access to model artifacts instantly.

Why This Is Optimal

This solution implements deep defense across network, identity, cryptographic, and operational logging layers. It strictly adheres to enterprise separation of duties and standard Google Cloud reference architectures for sensitive AI deployments.

View source documentation

Professional Cloud Architect

Secure Ai Model Deployment Vpcsc Cmek Iam

Professional Cloud Architect

Secure Ai Model Deployment Vpcsc Cmek Iam

What This Solution Is

How It Meets the Requirements

Key Advantages and Capabilities

Why This Is Optimal

Related questions

BigQuery Generative AI Functions