What happens when AI agents get the keys to your live network infrastructure?

CodiLime’s Solutions Architect, Tomasz Janaszka, PhD, explores the practical reality of connecting AI agents via the Model Context Protocol (MCP). In this session, he breaks down the security gaps that emerge when moving from human-operated to agent-operated networks and demonstrates how to build a multi-layered defense system.

It is a technical deep dive into securing real-world deployments using Keycloak, Open Policy Agent (OPA), and OpenBao, ensuring that agents act only within their authorized scope.

Moreover, Tomasz shows a live demo of Net-Inspector, a system showing how to enforce role-based tool filtering and dynamic maintenance windows that lock out AI agents in real time.

CodiLime builds network and infrastructure software for some of the most demanding environments in tech.