Judgment Telemetry for Agentic AI Systems: From Behavioural Events to Auditable Responsibility Flows

This working paper proposes a minimal judgment telemetry schema for agentic AI systems. While contemporary runtime governance frameworks focus on behavioural telemetry—actions taken, tools invoked, policies triggered—they systematically omit judgment delegation and responsibility flow. The schema introduced here specifies governance-facing fields that render delegation events, authority states, override conditions, and responsibility anchoring observable and auditable, without requiring access to model internals or proprietary decision logic.