The Judgment Infrastructure Gap Making Judgment Delegation Observable and Auditable in Agentic AI Systems

This working paper introduces a governance-level framework for auditing judgment delegation in agentic AI deployments. It reframes AI governance from a model-centric view (fairness, transparency, explainability at the level of individual systems) to a system-level view in which judgment, evidentiary weight, and responsibility flow across human, institutional, and machine nodes. The paper specifies delegation events, structural risk zones (judgment surrender, decision drift, responsibility fallback), and minimal auditability requirements necessary to restore accountability without access to proprietary model internals. It is written to be legible to policy and governance audiences while maintaining methodological safety boundaries.