AI is changing how applications communicate, execute actions, and interact with data. LLM applications, inference services, autonomous agents, model APIs, vector databases, and MCP-based tools introduce new communication patterns and security boundaries that traditional network security was not designed to address.

This session explores how network security must evolve as AI workloads become integral to cloud-native infrastructure. It examines the shift beyond IP, port, and application-based controls toward workload identity, application context, data sensitivity, mode and AI-specific risks.

The session maps security boundaries across inference, agent-to-agent communication, model APIs, MCP servers, external tools and explores practical approaches to securing these interactions.

Drawing on cloud and virtual network security experience, the talk proposes an architectural framework combining cloud-native networking, identity, application-layer policies, and AI-aware security controls.