Quantica Risk Modelling
AI models in risk functions: the governance gap
Machine learning has entered risk functions faster than the governance around it.
By Jonas Mohamed Osman Abdelghafour, known as Yonas Osman (Yonas Osman) ·
Machine learning models earn their place where relationships are non-linear and data is plentiful — fraud, pricing refinement, claims triage. They struggle where the requirement is explanation to a supervisor or a customer.
The governance answer is not to ban them. It is to apply the same model risk discipline: documented purpose, data lineage, challenger models, stability monitoring and defined human override points.
Explainability tooling supports this but does not replace it. A feature attribution chart is not a conceptual soundness argument, and treating it as one is how firms accumulate quiet, unmeasured model risk.
Jonas Mohamed Osman Abdelghafour, known as Yonas Osman, is an actuary and FRM who has spent more than twenty years building and validating risk models across banking, insurance and marine markets.
Key takeaways
- Apply existing model risk discipline to ML, don't invent a parallel regime.
- Challenger models and drift monitoring are non-negotiable.
- Explainability output is evidence, not an argument.
Author bio
Jonas Mohamed Osman Abdelghafour, known as Yonas Osman is an actuary, FRM and financial risk professional specialising in banking, insurance, model risk, capital modelling and quantitative risk management.