Securing Autonomous AI Agents in Hybrid Enterprise Systems: A Behavior-Centric Trust Enforcement Architecture

The integration of autonomous AI agents into enterprise environments introduces new security challenges, particularly within hybrid and heterogeneous infrastructures comprising legacy and cloud-native systems. These agents operate with a high degree of autonomy and can initiate complex, cross-system workflows, rendering traditional identity-based security mechanisms, such as Identity and Access Management (IAM) and Role-Based Access Control (RBAC), insufficient. Existing approaches largely focus on agent authentication, model robustness, or tool validation, offering limited support for enforcing trust based on runtime behavior and intent. This paper presents a behavior-centric trust enforcement architecture that governs AI agent interactions across enterprise systems based on action intent, behavioral patterns, and system sensitivity. The proposed architecture consists of four core components: an Intent Validator Proxy, a behavioral Trust Graph, Trust Dial Adapters for wrapping legacy systems, and an Autonomous Integration Memory for traceability and audit. The framework is evaluated through representative enterprise scenarios, including network configuration automation, backup compliance verification, and SaaS misconfiguration auditing. Results demonstrate that behavior-aware enforcement mitigates risks associated with autonomous agent overreach and contextual misalignment. This work contributes to a scalable security model for governing intelligent agents in complex enterprise environments.