# Implement Agentic Identity to Secure AI Agents with Kong AI Gateway

This talk explains how to implement agentic identity in Kong’s gateway today, with a practical focus on securing MCP, model, and agent traffic. Andrew Jessup argues that agents are unusually “chatty” systems that interact with many internal and external services, which creates identity sprawl, weak auditability, and security risk if agents hold too many direct credentials. The proposed approach is to put a gateway in the middle so every action is monitored, tied to both the agent and the human it acts for, evaluated against policy, and executed upstream with the correct tokens outside the agent itself. He outlines a staged path: first observe existing MCP traffic in pass-through mode, then establish inbound identity through OAuth and a dedicated MCP authorization server linked to corporate identity, then handle outbound identity with token pass-through, token exchange, or token vaulting. Finally, Kong policies, principals, groups, and CEL-based authorization can enforce fine-grained access, rate limits, and governance.