Governed autonomy for CSDs : A dual-layer AI framework built on an event-driven foundation

Global securities settlement faces ongoing friction from settlement fails, costing billions annually. While artificial intelligence (AI) promises a solution, grafting it onto legacy batch systems is pointless; AI cannot forecast effectively what it cannot see in real time. Drawing on operational practice with several central securities depositories (CSDs), this paper presents a foundation for meaningful modernisation. The paper suggests that shifting to an event-driven, in-memory architecture is the non-negotiable prerequisite for effective AI advancement. This foundation provides a dual-layer exception-handling approach where deterministic rules handle routine cases and a controlled AI layer tackles complex patterns. Through anonymised case studies, the paper quantifies achievable advantages, such as large reductions in settlement fails and resolution times and identification of hard limits tied to data quality. The framework includes governance mechanisms separating rule-based and AI-driven oversight, aligns with standards such as the National Institute of Standards and Technology (NIST) AI Risk Management Framework, and presents an 18–30 month risk-aware roadmap. This approach is described as augmented autonomy: a strategy centred on amplifying practitioner expertise, restricting automated actions with human oversight, and maintaining full audit trails to guide CSDs, exchanges, and regulators in deploying AI without sacrificing resilience or control. This article is also included in The Business & Management Collection which can be accessed at https://hstalks.com/business/.