Banks
NIST controls, model risk management, exam findings
- What the reviewer asks for
- Ongoing monitoring of the agent action, beyond a model inventory.
- What ICA hands them
- Runtime evidence of every consequential agent action, sealed and re-performable, so model risk can challenge the action in use, with the inventory kept as it is.
- Where to start
- One credit or operations path.Model risk or internal audit accepts the evidence package: yes or no.
- Beside what you run
- An inventory lists the model. ICA's receipt proves what the agent did, who authorized it, and that the record was never edited.
SR 26-2 (Federal Reserve, OCC and FDIC, 17 April 2026) supersedes SR 11-7, places generative and agentic AI outside its scope, and leaves each bank's own governance to set the controls for them. It is most relevant to banks above $30 billion in total assets.
SR 26-2 · checked 3 October 2026