Hong Kong launches ART for banking AI: suppliers must map data, access and human control

Important news | Hong Kong’s ART framework centres on accountability, resilience and trust. Suppliers should prepare controlled, reviewable use cases.

Editorial desk

Illustrative scene: training-data sharing workstation and source folders; no actual facility, interface, statistics, endorsement or logos.

On 15 October 2026 the Hong Kong Monetary Authority announced ART, an AI framework for banking; JETRO reported it on 6 November. ART stands for Accountability, Resilience and Trust and is intended to support AI adoption while preserving governance and operational resilience. It is a supervisory direction and support framework, not blanket approval for every agentic-AI service or permission to bypass existing risk processes.

The framework has three workstreams: supervisory guidance for responsible AI adoption; a training-data sharing platform with Cyberport and the University of Hong Kong; and further agentic-AI trials through a new sandbox phase. A software supplier pitching to Hong Kong banks should explain data provenance, tool permissions, task boundaries, human takeover and shutdown behaviour, not only demonstrate model answers.

HKMA said more than 60% of surveyed cases involved agentic AI, including the planning and execution of multi-step tasks within defined boundaries and controls. Banks are therefore exploring systems that touch workflows, not just chat interfaces. Providers of customer service, compliance, document-review or operations agents should map every input, permission, output and reviewer instead of simply promising efficiency.

Illustrative scene: multi-step agent workflow cards and permission boundary groups; no actual facility, interface, statistics, endorsement or logos.

Editorially, ART clarifies what suppliers should prepare but does not guarantee procurement. Start with a two-page banking brief: define one task, permitted and prohibited data and the accountable owner; then show failure cases, human review, recordkeeping and recovery. Seek a low-risk trial with a partner bank. Because a discussion paper on agentic-AI governance is still forthcoming, contracts and product roadmaps need room to adapt.

Illustrative scene: governance and failure-recovery brief with review records; no actual facility, interface, statistics, endorsement or logos.

Sources

Some images in this article are generated illustrations.