The Unpredictability Problem in AI
The rapid adoption of artificial intelligence agents and applications is creating significant challenges for enterprise security teams. Unlike traditional deterministic software, AI systems are non-deterministic, meaning their outputs cannot be precisely predefined. This unpredictability dramatically expands the potential blast radius of a security incident, while business pressure to deploy AI quickly often leaves security measures trailing behind. Niv Braun, co-founder and CEO of Noma Security, emphasized that this new reality demands a fundamentally different security approach.
Context and Runtime as the Foundation
According to Braun, effective AI security must be built on two core pillars: a flexible holistic framework that can incorporate rapidly evolving technologies such as the Model Context Protocol (MCP), and deep contextualization that fuses posture management, access controls, and runtime monitoring into a single coherent signal. He stressed that without visibility into runtime behavior, it is impossible to provide accurate recommendations for configuring agents or determining appropriate access levels. A unified AI security platform, he argued, is far more effective than relying on isolated point products to distinguish legitimate agent actions from genuine threats.
Source: Healthcareinfosecurity
