The Core Challenge of AI Unpredictability
The rapid adoption of AI agents and applications is creating significant security challenges for enterprises. Unlike traditional software, which is deterministic and predictable, AI systems are non-deterministic, meaning their actions cannot be fully anticipated. This fundamental difference, combined with the large potential blast radius of an AI compromise and intense pressure to deploy quickly, leaves security teams struggling to keep pace, according to Niv Braun, co-founder and CEO of Noma Security.
A Framework Based on Context and Runtime Monitoring
To address this, Braun argues that AI security must be built on two pillars: a flexible, holistic framework that can absorb fast evolving technologies like the Model Context Protocol (MCP), and deep contextualization that unifies posture management, access controls, and runtime monitoring. The key insight is that without visibility into what happens at runtime, security teams cannot provide useful recommendations on configuration or access permissions. A unified AI security platform, rather than isolated point products, is essential for distinguishing legitimate agent actions from real risks.
Source: Healthcareinfosecurity
