The discipline, rooted in banking regulation, of controlling the risk that a model is wrong or misused.
It came out of financial-services supervision and centers on validating models, documenting them, and monitoring them over their life. AI agents stretch the classic MRM playbook: an agent is not one static model but a changing system of models, prompts, tools, and policies, so validating it once before deployment does not tell you how it behaves months and many changes later.
Related: Agent Risk Management, EU AI Act, AI Agent Governance
See also: AI Agent Drift in Regulated Environments