PRA SS1/23 model risk for an LLM agent
Tier an LLM agent as a model, validate it, and produce the SS1/23 audit trail a UK bank's MRM function needs.
For: Model risk / SS1/23 validators in UK financial services
The regulated problem
A UK bank deploying an LLM agent for decisioning has to answer whether it is a 'model' under PRA SS1/23, tier it by materiality, and evidence ongoing validation and monitoring — inside an existing model-risk-management framework, with an audit trail the PRA can inspect.
How Regulus handles it
Tier the agent by materiality
The model-risk plugin classifies and tiers the LLM agent against SS1/23's materiality dimensions, so higher-tier agents route through stricter controls and dual-control on high-risk tools.
Consumer Duty outcomes monitoring
The FCA SYSC and Consumer Duty profiles compose alongside SS1/23, so outcomes-monitoring obligations for AI-driven decisioning are enforced at the same tool boundary rather than bolted on separately.
An audit trail the validator can quote
Every decision lands in a hash-chained envelope with the resolved clause and framework citation — the ongoing-validation evidence an MRM function and the PRA can reconcile, produced by the runtime itself.
Regulus is the open-source EU + UK compliance plane for Google ADK, a project by Neul Labs.