Agentic AI can help financial institutions respond faster, reduce costs and increase capacity. Greater autonomy does not transfer an institution’s supervisory accountability to the system. The practical question is whether the FI can justify the work it entrusts to AI and the arrangements used to control that work.
This article distinguishes continuing supervisory accountability from potentially divided contractual responsibility. It sets out six questions for assessing whether automation is reasonable, covering delegated authority, foreseeable harm, the scale and speed of errors, detectability and reversibility, safeguard reliability, and effective human intervention.
It also offers starting steps for implementing the Monetary Authority of Singapore’s AI risk management guidelines proportionately to the institution and its use cases.
