Agentic AI Drift Measurement and the Semantic Deviation Index (SDI): A Runtime Measurement Standard for the Governance of Agentic AI in Financial Services

Existing control frameworks in financial services share a structural limitation: they measure what agentic systems do, not what those systems understood before they acted. The failure originates before execution begins, at the layer where meaning is resolved, and is not observable through post-execution monitoring. This paper introduces the Semantic Deviation Index (SDI) as a runtime measurement standard that operationalizes governance at the reasoning layer of agentic AI systems deployed within financial services. The SDI measures the divergence between the definition an agentic system resolves at the orchestration layer and the definition the institution has authorized. It provides a mechanism to determine whether execution is governed before it occurs. The SDI is the measurement layer within a broader governance architecture introduced in prior work. Agentic Workflow Drift is the mechanism by which agentic systems pull inconsistent definitions from across the enterprise into a working logic no human authorized. Agentic Workflow Subversion is the risk surface that results when drift propagates unchecked or is deliberately exploited. The Semantic Control Plane is the enforcement architecture. The SDI operationalizes that architecture: an operational measurement mechanism that makes semantic deviation auditable and actionable in real time. This paper responds directly to the call made in Doyle-Spare (2026a), which introduced the taxonomy and called specifically for a semantic deviation index as the empirical anchor needed to advance the framework from conceptual taxonomy to measurable governance practice. The paper establishes that the SDI is not a metric within existing frameworks. It is a new measurement standard for agentic systems, one that sits alongside SR 11-7 and the NIST AI Risk Management Framework as a necessary complement, not a substitute. It further introduces the Deterministic Gate as the enforcement mechanism the SDI operationalizes, the Agentic Blast Radius as the containment boundary it defines, the Semantic Audit Trail as the governance record it produces, and the Doyle-Spare Agentic Governance Model (AGM) as the complete governance architecture these components form together. The paper draws on practitioner expertise in banking controls, model risk governance, and enterprise architecture, and is submitted as original practitioner-led research into an emerging and underexamined governance gap. The SDI formalizes pre-execution validation of semantic alignment in agentic systems through a measurable, reproducible divergence construct.