Agentic Networking: Securing AI Agents on Kubernetes

AI agents function like next-generation microservices, but their autonomous behavior and unique communication patterns present challenges and new security needs for existing cloud-native infrastructure. Prompt injections can exfiltrate PII to third-party tools, and poisoned tool responses can manipulate agent decisions. Kubernetes was not originally designed for the intricate and often unpredictable traffic patterns of A2A, agent-to-tool, and agent-to-LLM communication.

This session introduces "Agentic Networking" to adapt Kubernetes for this new reality. We will dive into the core challenges posed by AI-first protocols like MCP and A2A, which require a fundamental rethinking of traffic management, security, and governance.

We will present our work extending the Kubernetes Gateway API to provide well-governed, auditable agentic traffic, with gateway-level guardrails to further secure agents running on Kubernetes. Join us to explore the future of Kubernetes networking in the age of AI.