EP 220: Every AI Agent Needs Its Own Name

Artificial intelligence disrupts traditional cybersecurity by replacing deterministic systems—which rely on predictable rules—with probabilistic models that lack clear, traceable logic. This shift creates a critical vulnerability where control and data planes are inextricably mixed, turning standard threats into complex prompt injection risks. Organizations face an escalating supply chain crisis, particularly when utilizing opaque frontier models that lack transparency regarding training data. To mitigate these risks, security teams must implement hybrid architectures that utilize deterministic "harnesses" to monitor and constrain AI agents. Managing this landscape requires rigorous asset management and a focus on the "lethal trifecta": models with internet connectivity, access to privileged data, and the ability to process untrusted inputs. Guest Dan Stocker, an expert in AI security architecture, emphasizes that these controls are essential for maintaining accountability in an environment where AI agents increasingly operate with human-like autonomy.