As agentic AI systems move from experimentation to production, the security challenge is no longer just model risk, it’s runtime behavior. Autonomous agents introduce a dynamic control plane that can take actions, chain decisions, and interact across systems at machine speed. When something goes wrong, the blast radius can expand quickly and unpredictably. This roundtable will explore how CISOs are approaching containment in agentic environments, including governance of the agentic control plane, strategies for limiting unintended consequences, and new approaches to identity, policy enforcement, and observability. The discussion will focus on what it takes to safely operationalize autonomy without losing control.