Artificial intelligence deployment is outpacing traditional security measures, creating a need for robust platform security. Protecting proprietary model weights, defending against new threats like prompt injection, and ensuring regulatory compliance without hindering AI development are key challenges. Google Kubernetes Engine (GKE) offers a solution through a blueprint for AI workload security, consolidating Google Cloud services and GKE features. This blueprint addresses security across three critical layers of the AI stack. The infrastructure layer ensures a secure cluster baseline with Confidential GKE Nodes for hardware-attested confidentiality and Zero-Trust Networking with Workload Identity Federation and VPC Service Controls. Model security focuses on provenance and behavioral integrity, using k8s-aibom for visibility into models, datasets, and frameworks. Application security defends the inference path with Model Armor to inspect prompts and responses for threats. Session management via the GKE Inference Gateway enforces quotas and detects abuse. Agentic isolation is achieved through GKE Sandbox, preventing container escapes and protecting nodes from unpredictable agent behavior. A phased approach to security, from deployment to operation and governance, is recommended. This comprehensive strategy ensures AI deployments are secure by default and scalable. Taking these best practices will prevent AI deployment from compromising security.