On 28 July 2025 the ECB published a revised Guide to internal models that, for the first time, adds a section on the use of machine-learning techniques in the 'Overarching principles' chapter. The guide requires that ML-based internal models be adequately explainable and that their performance justify their complexity, and it defines ML as highly complex, many-parameter techniques capturing non-linearity and needing large training datasets — explicitly excluding linear and logistic regression. It applies to all ECB-supervised banks using internal models for credit, market or counterparty credit risk.
| Document | ECB Guide to internal models (July 2025, ML section) — Revised ECB Guide to internal models — expectations for machine-learning techniques in internal models |
| Issued by | European Central Bank — Banking Supervision (SSM) |
| Type | Guidance |
| Status | In force |
| Published | Jul 28, 2025 |
| Effective | Jul 28, 2025 |
| Applies to | Significant institutions using internal models (credit, market, counterparty credit risk) to compute risk-weighted assets, including any machine-learning components. |
| Official source | bankingsupervision.europa.eu ↗ |
| Use cases | Model risk management · Credit scoring & underwriting |
What are the key points of ECB Guide to internal models (July 2025, ML section)?
- New ML section in the Overarching principles chapter, answering an industry request for clarification
- Two core tests: models must be adequately explainable, and added complexity must be justified by performance
- ML defined as complex, many-parameter, non-linear techniques trained on large (sometimes unstructured) datasets; linear/logistic regression are not ML for the guide
- Applies to internal models for credit risk, market risk and counterparty credit risk RWAs
- Complements the EBA's 2023 follow-up report on machine learning for IRB models
- The ECB's 2025 supervisory annual report cites the ML section as the year's key internal-model clarification
What did ECB Guide to internal models (July 2025, ML section) change for banks?
Until 2025 banks had no ECB statement on whether ML could sit inside an approved capital model. The revised guide says yes, on conditions — explainability and justified complexity — so ML components now enter the standard model-approval and ongoing-monitoring process rather than being a grey area.
Can euro-area banks use machine learning in IRB models?
Yes, since the July 2025 revised Guide to internal models, provided the model is adequately explainable and the complexity is justified by performance; ML components go through the ECB's internal-model approval process.
Is logistic regression 'machine learning' under the ECB guide?
No. The guide excludes linear and logistic regression from its ML definition.
| Date | Document | Status |
|---|---|---|
| Jul 7, 2026 | ECB 'Dear CEO' letter on AI-enabled cybersecurity threats (SSM-2026-0301) — Addressing AI-enabled cybersecurity threats — letter from the Chair of the Supervisory Board to CEOs of significant institutions | In force |
| Jun 3, 2026 | Elderson speech: 'Strengthening operational resilience for the age of AI' (June 2026) — Strengthening operational resilience for the age of AI — speech by Frank Elderson | Final |
| Feb 24, 2026 | Machado speech: 'Technology is neutral, governance is not' (Feb 2026) — Technology is neutral, governance is not: AI adoption in the banking sector — speech by Pedro Machado | Final |
| Feb 3, 2026 | Montagner speech: 'Encouraging innovation, managing risks' (Feb 2026) — Encouraging innovation, managing risks: the ECB's approach to digital transformation — speech by Patrick Montagner | Final |
| Nov 20, 2025 | Supervision Newsletter: AI use cases for credit scoring and fraud detection (Nov 2025) — AI's impact on banking: use cases for credit scoring and fraud detection (Supervision Newsletter, November 2025) | Final |
| Nov 18, 2025 | SSM supervisory priorities 2026–28 — ECB Banking Supervision: SSM supervisory priorities for 2026–28 | In force |
Follow every document these regulators publish
6 curated AI stories for banking executives · Every morning · Free
Subscribe to BankingNewsAI →