PRA News

Key dates

24 November 2025

- PRA published roundtable summary and slides. https://www.bankofengland.co.uk/prudential-regulation/publication/2025/november/pra-holds-model-risk-management-roundtable-on-ai

20

22 October 2025; - PRA held CRO roundtable sessions with 21 firms on AI/ML MRM

Suggested considerations

• Review and strengthen board-level model risk appetite statements to explicitly cover AI/ML opacity and uncertainty; integrate into governance triggers like re-validation.
• Enhance model inventories for completeness, aggregate risk assessment, and cross-jurisdictional tiering challenges.
• Update model development policies to evaluate AI/ML trade-offs (e.g., explainability vs. performance) and ensure datasets prevent overfitting.
• Revise ongoing monitoring policies for more frequent, quantitative checks on AI/ML (e.g., beyond six months); define degradation triggers, fallback models, and kill switches.
• Participate in PRA initiatives like MRM roundtables or AI Consortium for dialogue; align first/second-line defenses per SS1/23.

What changed

• This is not a formal rule change but supervisory guidance via roundtable insights reinforcing SS1/23 principles (effective since 2023). Key emphases include:
• Risk appetite: Boards must articulate AI/ML-specific model risk appetite pre-deployment to avoid exceeding tolerances, given higher uncertainty from opacity.
• Model inventories and tiering: Address inaccurate/incomplete inventories and aggregate risks from deploying similar AI/ML across portfolios/jurisdictions; challenge tiering for complexity.
• Model development: Assess trade-offs in performance vs. explainability/reliability; prefer simpler models where AI/ML gains are marginal; mitigate overfitting via representative datasets.
• Ongoing monitoring: Increase frequency beyond tier-dependent intervals (e.g., six months may suffice for traditional models but not dynamic AI/ML); define quantitative triggers for re-validation.