Die Session

Who Are You, Agent? — Identity & Authorization in the Age of Agentic AI

AI agents are rapidly evolving from chatbots that generate text into autonomous actors that call APIs, invoke tools through MCP, access enterprise data, and perform actions on our behalf.

But this creates a surprisingly difficult question: Who are we actually authorizing? Is it the human user, the AI agent, the workload running the agent — or all of them?

Giving an agent a powerful API key or service account might be convenient, but it creates a new class of highly privileged machine identities whose behavior is partly controlled by an LLM.

In this talk, we'll explore how established identity concepts and standards such as OAuth 2.1, OpenID Connect, Cross Domain Application Access, token exchange, workload identity, SPIFFE and AuthZEN can be applied to the emerging world of agentic AI.

We'll look at authorization for MCP-based tools, delegated access, least privilege, and the important distinction between an agent's own identity and the identity of the user on whose behalf it acts.

We'll also examine emerging approaches to Agent Identity, including Google's SPIFFE-based Agent Identity model, and derive a practical architecture for securely connecting users, agents, workloads, and APIs.