What is Model Risk Management (MRM) for AI?
Model risk management is how an organisation controls the risk of decisions based on incorrect or misused models: how models are developed, validated, documented and monitored, and who is accountable for them.
The current US guidance
In April 2026 the Federal Reserve, the OCC and the FDIC issued revised guidance on model risk management (published as SR 26-2), replacing SR 11-7 (2011). It takes a risk-based approach tailored to a banking organisation's model risk profile, and defines a model as a complex quantitative method that applies statistical, economic or financial theories to produce quantitative estimates.
What it excludes
Its definition excludes deterministic rule-based processes, and it places generative and agentic AI models outside its scope, while saying that a banking organisation's own risk management should still decide the controls for them.
Where NOMOS fits
NOMOS rules are deterministic: the same facts against the same sealed version always give the same verdict, and each verdict names its rule. Where a model's output feeds a decision, NOMOS holds the rule that acts on it, and the Registry replays recorded decisions against any new version before it takes effect.
Example
A bank uses a model to score fraud risk and a threshold to block transactions.
The threshold is changed in a configuration file. The model is validated; the threshold is not recorded anywhere a reviewer looks.
The threshold is a sealed rule. A change to it is a new version, reviewed with a replay of past decisions showing which would have been blocked differently.
Frequently asked questions
Related concepts
See it in practice
Upload a policy, confirm its rules, seal them: any system can then check an action against them.