This working paper identifies and defines the Judgment Infrastructure Gap: a structural absence in current AI governance frameworks whereby the delegation of judgment to AI systems is neither explicitly authorized, nor systematically observable, auditable, or reversible. Across high-stakes decision workflows—particularly in government, healthcare, and education—AI systems increasingly shape final decisions without corresponding governance mechanisms that track when, how, and by whom judgment authority is transferred. Existing approaches to AI governance, including principles-based guidelines and model-centric risk management frameworks, primarily focus on model behavior, outputs, or compliance processes. They do not adequately address judgment delegation as an operational governance problem. This paper reframes AI governance failures as infrastructure-level issues rather than model errors or ethical shortcomings. It proposes a conceptual and design-oriented framework for an observer-facing judgment middleware: a governance layer that makes judgment delegation visible and auditable within institutional decision workflows, without disclosing internal model mechanisms. The contribution of this work is threefold. First, it formally defines judgment delegation as a distinct object of governance. Second, it introduces observer-facing markers and state-flow risk trajectories that function as early-warning indicators for governance failure. Third, it positions judgment infrastructure as a missing but necessary complement to existing AI risk management, policy, and infrastructure frameworks. This paper is released as a public working paper to establish conceptual priority and to support further research, policy development, and systems design. It is not a full system disclosure, an empirical benchmark study, or a policy recommendation document.
Sincere Ann Ma (Tue,) studied this question.