Inside Sibos | Governance & Assurance | 2

Inside SIBOS | Governance & Assurance week | 2 Four controls to productionise AI for AML/sanctions-without drama. Why it matters: Controls before code = fewer...

Inside SIBOS | Governance & Assurance week | 2
Four controls to productionise AI for AML/sanctions-without drama.
Why it matters: Controls before code = fewer RFIs and smoother audits.
* Data gates: Block/route‑down when , , name/address, and LEI fill‑rates fall <95%; alert owners by corridor.
* Model dossier: Versioned model cards (purpose, features, training data), champion-challenger, drift thresholds; independent validation.
* Explainability & override: Ship per‑decision reasons; record human overrides with evidence; sample QA weekly.
* Privacy‑by‑design: Apply FEAT (Fairness, Ethics, Accountability, Transparency); log lineage for every dataset/feature.
* Third‑party resilience: Audit rights, reproducibility, and exit plans in contracts; test failover quarterly.

Game‑theory: Safety advocates now hold the pen (PRA/MAS set expectations). Quiet winners: firms proving auditability at scale-outpacing rivals still "exploring."

Topics: ISO 20022, SWIFT, Compliance, AI, Data Quality