Skip to content
Conference Open access

Toward Trustworthy Enterprise AI Agents: A Human-Governed Runtime Architecture for Secure Execution

Sep 2026 · 2026 International Conference on Cybersecurity and AI-Based Systems (Cyber-AI) · pp. 514-519 · 0 citations · 22 references

Abstract

Modern enterprise AI agents increasingly operate across external systems, APIs, and sensitive organizational workflows, introducing significant risks related to unauthorized actions, unreliable execution, and limited operational transparency. While large language models enable powerful autonomous capabilities, many current agent architectures lack sufficient governance, runtime control, and human oversight mechanisms required for production deployment in security-sensitive environments. This paper presents a human-governed runtime architecture for secure enterprise AI agents that separates decision generation from execution authorization through a controlled approval and observability layer. The proposed architecture incorporates human-in-the-loop validation, scoped execution permissions, runtime policy enforcement, structured auditability, and execution isolation to reduce operational and security risks associated with autonomous agents. We further introduce runtime observability mechanisms for tracking agent plans, execution outcomes, and approval workflows across integrated enterprise services. A prototype implementation integrating external developer and productivity platforms demonstrates how governance-driven runtime orchestration can improve trustworthiness, operational accountability, and execution safety without eliminating the flexibility of AI-assisted automation. Experimental evaluation and architectural analysis illustrate the effectiveness of the approach in reducing unsafe autonomous actions while maintaining practical usability for enterprise environments.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.