Why the existing frameworks fall short
SR 11-7 and equivalent frameworks were built for statistical and credit models. They map only loosely onto retrieval-augmented LLM systems. This whitepaper closes the gap with a complementary set of controls.
Whitepaper · Risk
Model risk management for the GenAI era
SR 11-7 was written for credit models. Here is how to extend it for LLMs your regulator will accept.
22
Pages
11 min
Read time
Risk
Topic
PDF
Format
Key takeaways
- 01 A field-tested extension of SR 11-7 controls to LLM-based decisioning.
- 02 Templates for model risk records, evaluation harnesses, and kill-switches.
- 03 Regulator engagement protocol that has earned first-attempt acceptance in three jurisdictions.
A complementary control set
Eight controls covering input handling, retrieval verification, output policy enforcement, citation enforcement, evaluation harness operation, drift detection, kill-switch protocols, and post-hoc audit reproducibility.
Templates that work
Model risk records, kill-switch playbooks, and regulator-facing artefacts available for adoption — all proven in production with central banks and equivalents.
Working with the regulator
A pre-tested engagement protocol covering pre-briefings, demonstration scripts, and evidence packs. Used in three jurisdictions with first-attempt acceptance.
More research
All whitepapers →
Related whitepapers.
Architecture
Sovereign LLM platforms
14 minSovereign LLM platforms — a reference architecture
Read whitepaper
Modernisation
The strangler-fig playbook for core banking
16 minThe strangler-fig playbook for core banking
Read whitepaper
FinOps
FinOps reference: from cloud waste to commitment optimisation
9 minFinOps reference: from cloud waste to commitment optimisation
Read whitepaper
Download
Get the full whitepaper.
PDF available to enterprise subscribers. We’ll route the request to the right partner for a follow-up.