Loading page…
Compliance Week’s October 24 coverage highlights proportionate risk controls alongside governance. The framework also recognises oversight without human involvement at every step.
Listen to this story
Under the framework issued October 15, 2026, banks are expected to assign ownership to each AI use case, tie agent controls to a defined purpose, map system dependencies and prepare ways to intervene or recover. The guidance reflects the HKMA’s sandbox findings: more than 60% of tested use cases involved agents, and half of those used multiple agents. A proposed supervisory paper will consider a registry for material agents; that registry is not yet a requirement.
Carmen Chu said oversight need not involve humans at every step, but must include clear accountability and ongoing monitoring.
Banks should map dependencies across models, data, tools, infrastructure and third-party providers, and test behavior under extreme scenarios.
Possible safeguards include kill switches and behavioral limits that reflect a bank’s risk appetite.
Hong Kong’s updated bank AI guidance centres on three priorities: clear governance, proportionate management of AI risks, and trust and transparency, Compliance Week reported October 24. The Hong Kong Monetary Authority issued the guidance with its Accountability, Resilience and Trust framework on October 15, 2026, setting expectations for banks adopting increasingly autonomous systems.
The framework draws on the regulator’s second generative AI sandbox cohort, a programme for testing applications. The HKMA said more than 60% of its use cases involved AI agents; half of those involved multiple agents. These systems can plan multi-step tasks, use tools and access data and systems—not just generate recommendations.
In October 16 remarks explaining the framework, Carmen Chu, the HKMA’s executive director for banking supervision, said each AI use case should have clear ownership. Banks should also maintain an up-to-date inventory of approved applications.
For agents, Chu said accountability should be anchored to intent: banks should define what an agent is meant to achieve. That purpose should carry through its design, data, permitted tools and available actions. The guidance recognises a shift toward people supervising a system rather than participating in every step.
As A.I. use cases become more sophisticated, effective oversight need not mean human involvement in every step. But it must mean clear accountability and on-going monitoring.
Carmen Chu, HKMA executive director for banking supervision, October 16, 2026
Resilience begins with knowing what supports an AI application. Chu said banks need to map connections across models, data, tools, infrastructure and third-party providers. They should have response and recovery capabilities, with measures ready to correct or restrict a system.
Alongside the issued guidance, the HKMA plans a discussion paper on a proposed supervisory framework for agentic AI governance. It will explore accountability tied to an agent’s intended purpose, a registry of material AI agents and effective intervention mechanisms.
The framework also includes a planned training-data sharing platform with Cyberport and the University of Hong Kong, with quality assurance built into its design. Chu linked the initiative to a sandbox lesson: effective validation depends on quality data. The platform is intended to support both training and testing.
Loading discussion...
Join the conversation
Explain what would make that accountability meaningful to you.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.