The Unpredictability Problem in AI Security
Traditional software operates on deterministic logic: it reliably executes predefined instructions. AI applications shatter this predictability, introducing a non deterministic behavior that expands the potential blast radius for security incidents. According to Niv Braun, CEO of Noma Security, this fundamental shift creates immense pressure on enterprise security teams who must balance rapid AI deployment with the need to contain unknown risks. The technology’s inherent unpredictability means standard security approaches are no longer sufficient.
A Framework Built on Context and Runtime Awareness
To address these challenges, Braun advocates for a security strategy based on two core pillars: a flexible holistic framework capable of absorbing fast evolving technologies like the Model Context Protocol (MCP), and deep contextualization that merges posture management, access controls, and runtime monitoring into a single coherent signal. Without visibility into what occurs during runtime, security teams cannot offer effective configuration recommendations or determine appropriate access privileges for AI agents. A unified AI security platform is essential, as siloed point products fail to provide the comprehensive view needed to distinguish legitimate agent actions from genuine threats.
Impact and Scope
This approach is critical for organizations racing to adopt AI without letting security lag behind. By emphasizing context and runtime awareness, security teams can better manage the enormous risk surface introduced by AI agents. Braun, who previously led security in Israeli intelligence and helped shape AI security standards, stresses that early partnerships between AI providers and security vendors can enable secure by design capabilities from the outset. As AI proliferation continues, a unified strategy that connects policy, access, and real time monitoring will be key to protecting enterprise assets.
Source: Healthcareinfosecurity
