The ECB just raised the bar
Model Risk Management is now expected for all models — not just the regulatory ones.
At the end of July, the European Central Bank (ECB) published its latest Supervisory Guide on Internal Models (2025). The new guidance brings a big shift for financial institutions across the EU: Model Risk expectations are now clearly formalized — and no longer confined to internal models under regulatory scope. They now extend to every model used by your institution. That means pricing models, stress testing models, ALM simulations, and yes — AI and ML models too.
Why this matters
Until now, many institutions maintained a two-tier model governance setup: strong controls around regulatory/internal models (think Basel IRB models), and more lightweight—or even ad hoc—controls around everything else. That approach is no longer good enough.
“A comprehensive model risk management framework is expected to apply to all models… regardless of their regulatory classification.”
– ECB Supervisory Guide 2025, Chapter 4.1
In other words, if a model supports decision-making or exposes the institution to risk, it falls under the MRM umbrella. Full stop.
This shift reflects a broader trend: as models become more complex and business-critical (especially with the rise of AI), the risks associated with poor governance are no longer confined to just regulatory areas. Think: pricing errors, reputational damage, biased decisioning, or opaque automation.
A spotlight on ML: welcomed, but scrutinized
In fact, the guide introduces a brand new chapter (Chapter 9) focused specifically on machine learning models. The tone is constructive: the ECB welcomes the use of ML-based techniques — but warns that they must meet higher standards than traditional models.
Specifically, ML models are expected to:
- Be fully traceable, versioned and explainable
- Include clear documentation of features, training data, and architecture
- Be subject to independent validation, at least annually
- Be supported by auditable decision logs, especially in production
“The IT infrastructure should allow an institution to replicate and verify the decisions made by ML-based models.”
– ECB, Chapter 9, Paragraph 53
For many institutions, this will require a rethink. A notebook is no longer enough — regulators now expect auditability, transparency, and traceability by design.
How Yields helps you stay ahead
At Yields, we’ve been preparing for this shift. Our platform was built specifically to support enterprise-grade Model Risk Management — across all model types including AI models and Systems.
With Yields, you can:
- Maintain a centralized model inventory
- Capture version histories, training data, approvals and ownership
- Run structured validation workflows
- Generate audit-ready documentation and reports — at any time
- Support explainability, even for black-box models
Whether you’re dealing with internal regulatory models or cutting-edge third-party LLMs, we help you bring order, governance and control — without slowing innovation.
The takeaway
The ECB has made it clear: every model matters. And the institutions that take this seriously will not only avoid compliance issues — they’ll also be better positioned to deploy AI safely, at scale, and with confidence. Start from the foundations. Implement effective model risk management.
Curious how your MRM setup stacks up? Book a demo or reach out — we’re happy to help.
