The Sentinel System: A First Operational Architecture for External, Trajectory-Aware AI Oversight

Artificial intelligence is entering a phase in which its capabilities increasingly exceed the mechanisms available for oversight and control. As systems become more autonomous, adaptive, and integrated into critical infrastructures, the need for external, real-time governance becomes essential. This work presents the first operational architecture of the Sentinel System, an external AI oversight framework designed to monitor, evaluate, and regulate model behavior independently from the systems it supervises. The proposed system introduces a structured decision layer capable of analyzing proposed actions, assigning dynamic risk levels, and enforcing outcomes including allow, modify, and block. Unlike traditional approaches based on isolated step-wise evaluation, Sentinel incorporates trajectory-aware analysis, enabling the detection of behavioral evolution across sequences of interaction. Through a controlled experimental framework, the system is evaluated across both isolated inputs and multi-step behavioral scenarios. Results show that while step-wise evaluation can correctly classify individual actions, it fails to capture progressive escalation patterns. The trajectory-aware extension enables early detection of behavioral drift and allows intervention before critical thresholds are reached. The paper also introduces a minimal formalization of this approach, including risk aggregation over time and a decision policy based on cumulative behavioral context. These elements establish the foundation for a new class of AI oversight systems capable of interpreting behavior as a continuous process rather than discrete events. This work represents a first step toward a scalable, verifiable, and operational infrastructure for AI governance.