On 24 November 2025 the PRA published the slides from two Chief Risk Officer roundtables held on 20 and 22 October 2025 with 21 PRA-regulated firms on applying the SS1/23 model risk management principles to AI and machine learning. The PRA set out its thinking on risk appetite for AI models, model tiering, explainability and interpretability, data and overfitting, independent validation, and ongoing monitoring — the closest thing to AI-specific model-risk guidance the PRA has issued.
OFFICIAL TEXT: bankofengland.co.uk ↗ · FINAL · UK (BOE / PRA / FCA)
| Document | PRA AI/ML model-risk roundtable (Nov 2025) — The PRA holds model risk management roundtable on artificial intelligence and machine learning technologies |
| Issued by | Bank of England, Prudential Regulation Authority, and Financial Conduct Authority |
| Type | Guidance |
| Status | Final |
| Published | Nov 24, 2025 |
| Applies to | PRA-regulated banks applying SS1/23, particularly those with internal-model approvals |
| Official source | bankofengland.co.uk ↗ |
| Use cases | Model risk management · Generative & agentic AI · Credit scoring & underwriting · Third-party & vendor AI |
What are the key points of PRA AI/ML model-risk roundtable (Nov 2025)?
- Two CRO roundtables, 20 and 22 October 2025, 21 firms; presentation slides published on the PRA site.
- Frames SS1/23 as the supervisory expectation for AI/ML models, not a separate regime.
- Topics: risk appetite, model tiering and materiality, explainability and interpretability, data quality and overfitting, independent validation, ongoing model monitoring.
- Cited by the BoE/PRA in both 1 April 2026 letters as evidence of proactive AI supervision.
- The PRA said it intends to build further on SS1/23 for AI in 2026 using these supervisory insights.
What did PRA AI/ML model-risk roundtable (Nov 2025) change for banks?
It signals how PRA supervisors will test AI models against SS1/23 in the 2026 supervisory cycle, where AI adoption is a named priority. Banks should expect questions on tiering of generative and vendor models, validation of non-deterministic outputs, and monitoring for drift.
Has the PRA issued guidance on AI under SS1/23?
Not formal guidance, but in November 2025 it published roundtable slides setting out its views on risk appetite, tiering, explainability, data, validation and monitoring for AI/ML models under SS1/23, and has said it will build on SS1/23 in 2026.
How does PRA AI/ML model-risk roundtable (Nov 2025) compare?
- PRA SS1/23 vs SR 26-2 — UK and US Model Risk Compared
| Date | Document | Status |
|---|---|---|
| Sep 30, 2026 | Bailey: Frontier AI and the Question of Governance (Sep 2026) — Frontier AI and the Question of Governance — Governor Andrew Bailey | Final |
| Sep 2, 2026 | FCA multi-firm review: Frontier AI and cyber resilience (Sep 2026) — Frontier AI and Cyber Resilience | In force |
| Jul 14, 2026 | HM Treasury Financial Services AI Adoption Plan (Jul 2026) — Financial Services AI Adoption Plan | Final |
| Jun 5, 2026 | 2026 BoE/FCA AI survey — The Bank of England and FCA's 2026 AI Survey | Final |
| May 15, 2026 | BoE/FCA/HMT joint statement on frontier AI and cyber resilience (May 2026) — The Bank, FCA and HM Treasury joint statement on Frontier AI models and cyber resilience | In force |
| Apr 1, 2026 | BoE response to Treasury Committee AI inquiry (Apr 2026) — Response to TSC inquiry report on AI in financial services | Final |
Follow every document these regulators publish
when one of these regulators moves, the next morning's brief says so · six sourced stories · 7 am ET · free
plus every tracker, bank and agent page update, the morning after · leave any morning