When AI Can Act, Accountability Needs an Owner

On 15 October, the Hong Kong Monetary Authority announced its “ART” framework for responsible AI adoption in banking. It includes updated supervisory guidance and further sandbox testing; a supervisory framework for agentic AI governance is still proposed for future discussion, not a finalized policy. Read the HKMA announcement.

The direction is useful: as AI systems gain access to data and workflows, governance needs to address not just what a system may do, but who owns the decision when it reaches a boundary. My practical test would be to name the authorized scope, accountable decision owner, escalation trigger, and review path for each consequential task. That is my interpretation, not an HKMA requirement.

When an AI system’s action affects people, who should be able to pause it—and who is responsible for explaining the trade-off?