Rethinking identity, permissions and trust in the age of agentic AI

AI agents are rapidly moving from assistants that suggest code to autonomous actors capable of calling APIs, modifying infrastructure, interacting with CI/CD pipelines, and accessing cloud environments.

This creates a new security problem: our next privileged user may not be human.

In this session, we'll explore how agentic AI changes the traditional identity and trust model. Through practical cloud and DevSecOps scenarios, we'll examine what happens when AI agents receive access to tools, credentials, pipelines, and production infrastructure, and what can go wrong when those permissions are poorly designed.

We'll look at practical approaches for securing agent identities using least privilege, short-lived credentials, workload identity, policy enforcement, human approval boundaries, and auditability.

Attendees will leave with a practical security model for answering a question that every engineering organization adopting AI agents will soon face: what should an AI agent actually be allowed to do?