Security | Threat Detection | Cyberattacks | DevSecOps | Compliance

Securing AI agents in Teleport, focused on unified identity, eliminating standing privileges, and enforcing real policy controls instead of relying on the whims of an agent.

Regulators don’t just want login logs anymore. They want immutable proof of every action and full session recordings. Steven Martin on how Teleport delivers that — and how AI-powered session summaries are changing what audit actually looks like.

There’s a moment in every Teleport demo where the customer lights up. They see session replay, Kubernetes playback, the scope of what’s possible — and something clicks. Gus Luxton on why that reaction never gets old.

AI agents are tireless, highly capable, eager to please, but difficult to manage. George Chamales (CriticalSec) and Josh Rector (Ace of Cloud) unpack the identity and access challenges posed by agentic AI. How do you verify it was the right agent, doing the right action, approved by the right person? How do we bound, constrain, govern agentic behavior? Ultimately, the same frameworks built for human identity and access should be applied to agents.

FedRAMP 20x → continuous monitoring. Moving from point-in-time audits to real-time visibility.

With just a Slack message, AI agents can triage and resolve Kubernetes failures. But what's stopping that agent from wiping your cluster?

In this video, we demonstrate how to securely grant an AI agent (OpenClaw) access to Teleport-protected Kubernetes resources using Teleport Machine Identity and tbot, without exposing secrets, API keys, or long-lived tokens. You’ll see how Teleport treats AI agents as first-class identities, enforcing strict RBAC controls so the agent can only do what it’s allowed to do, like reading logs, while being blocked from sensitive actions like deleting resources or accessing secrets.

SOC 2 was built for human-operated systems and predictable infrastructure. Agentic AI introduces systems that can act, adapt, and change over time, creating new governance and risk considerations.

Three methods for issuing identity to AI agents — and why static credentials will always eventually leak no matter how well you vault them. Ev Kontsevoy breaks down standard credentials, durable identity, and digital twins, and explains why the issuer of identity needs to be the same across your entire environment.

Most organizations have identity over here and infrastructure over there — and they don't talk. By default, infrastructure has no identity. It's naked. Ev Kontsevoy explains why bringing identity into your infrastructure stack is a prerequisite for safe AI adoption — and what a trusted state actually looks like.