Non-deterministic agents break the rules traditional software security was built on
AI agent security has become the critical gating factor for enterprises that want autonomous software doing real work. Unlike chatbots that answer questions, agents take actions, and every tool or credential they touch widens the potential damage. To that end, Red Hat Inc. has spent the past year building guardrails for that shift, including a […] The post Non-deterministic agents break the rules…
Recent developments in AI agent security have emerged as a critical concern for enterprises seeking to deploy autonomous software capable of performing complex tasks. Unlike traditional chatbots, these agents can actively take actions, making their potential impact far greater. Red Hat Inc. has been working on building safeguards for this new landscape, including a sandbox for OpenClaw agents.
Joe Fernandes, Red Hat's vice president and general manager of AI, pointed out that agents differ from traditional software in that they are non-deterministic. Rather than performing actions as programmed, they take actions to achieve a specific goal, making their behavior unpredictable. During a recent KubeCon + CloudNativeCon NA event, Fernandes and Jessica Forrester, Red Hat's senior distinguished engineer and chief architect of AI, discussed the various security measures Red Hat is implementing for autonomous AI agents.
These measures include sandboxing, identity management, access controls, and enterprise-scale agent management. To address the issue of agents potentially gaining access to sensitive credentials, Red Hat uses a technique called masked authentication. In this approach, agents are given access to the necessary tools while their sensitive keys are masked and replaced on the way out.
OpenShell, an open-source agent runtime developed by Red Hat and Nvidia Corp., allows agents to operate with limited privileges, ensuring they don't have actual access to critical resources. Red Hat has also implemented kernel-level agent isolation and identity management, assigning each agent its own identity and narrow permissions set.
By integrating with various identity systems, Red Hat can assign agent identities and authorize what actions they can and cannot perform. Looking ahead, Red Hat is working on OpenClaw Enterprise, a solution for managing thousands of personal agents at scale. Fernandes emphasized that the project has reached a stable milestone, indicating that it could be considered reliable for enterprise production use cases.
Written by urgent.news from SiliconANGLE's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
