Risk Dimensioning and Dispositioning: A Control-Based Production-Enablement Methodology for Governing Machine Learning, Generative, and Agentic AI in Financial Institutions

Authors

  • Sandeep Kumar Anuguthala

Keywords:

AI governance, machine learning risk management, generative AI governance, agentic AI risk, model risk management, first-line-of-defence, risk dimensioning, risk dispositioning, production-enablement, technology risk controls, financial institutions, AI regulatory compliance, residual risk assessment, control posture validation, enterprise AI governance, responsible AI deployment, operational risk, AI risk framework, pre-production risk gate, AI oversight

Abstract

Banks operate under regulatory obligations that require every risk introduced by a new system to be identified, measured, and formally accepted before that system touches customers, data, or decisions. Artificial intelligence is not exempt. It heightens the obligation, because AI systems behave probabilistically, drift over time, inherit bias from data, and can act autonomously in ways that are difficult to reverse. Without a formal gate, the production decision becomes a business decision made without a complete risk picture. This paper presents risk dimensioning and dispositioning, a control-based production-enablement methodology operated inside a first-line-of-defence technology risk and control function at a large financial institution between early 2024 and mid-2026. It takes a use case through six sequenced steps: intake and risk scoping, inherent risk identification, control posture validation, residual risk measurement and dispositioning, risk-informed decision facilitation, and non-objection and second-line approval facilitation. Dimensioning decomposes the system's risk profile across distinct domains and asks what exposure remains once controls are accounted for, rather than whether a control exists. Dispositioning closes the assessment into a documented production decision in which the use-case owner accepts residual risk, the second line of defence provides non-objection rather than co-approval, and Legal clears regulatory and contractual exposure, all three required before production, with the recurring discipline that the practitioner sights the artefact rather than accepting an assertion. The paper develops three of the framework's five risk dimensions, model risk, data risk, and technology controls, the third to the level of its control domains and verification standard. It reports the operating record of the programme the methodology ran inside, separately from the author's personal scope of 15 use cases led end-to-end through all six steps. No efficiency, speed, or cost claim is made, because no instrumented baseline exists.

Downloads

Published

2026-09-28

How to Cite

Anuguthala, S. K. (2026). Risk Dimensioning and Dispositioning: A Control-Based Production-Enablement Methodology for Governing Machine Learning, Generative, and Agentic AI in Financial Institutions. International Journal of Artificial Intelligence and Machine Learning, 6(12s), 1649–1670. Retrieved from https://www.svedbergopen.com/index.php/ijaiml/article/view/3176