Christopher Kelley
AI Governance / Model risk

What model risk management borrows from banking.

SR 11-7 as a template, and its limits for generative systems.

By Christopher M. KelleyEssay 7 min Apr 2026 5 sources
AI Governance · Model risk

Banking spent two decades learning to govern models it could not fully explain. That inheritance is worth taking — and worth knowing where it stops.

What transfers

Independent validation, an inventory of every model in use, and the discipline of treating the model as a liability until proven otherwise. None of that is new; all of it is missing from most AI programs.

Where the analogy breaks

A credit model has a narrow, stable output space. A generative system does not. The old validation playbook assumes you can enumerate what the model will be asked to do — and that assumption is exactly what breaks. The transferable part is the posture; the mechanics have to be rebuilt for open-ended output.

← All writing in AI Governance