What does HKMA "ART" Framework / AI supervisory guidance say about AI in banking?

What are the key points of HKMA "ART" Framework / AI supervisory guidance?

• Three pillars: Accountability (governance tied to an AI agent's intended purpose, carried through design, data, tools and actions — a 'prerequisite for effective intervention'), Resilience (mapping AI dependencies across models, data, tools, infrastructure and third parties, with response/recovery measures that may include a 'kill-switch'), and Trust (securing the AI agent's operating boundaries, tested under extreme scenarios, resting on data quality).
• Oversight: the guidance 'recognises a progressive shift from "human-in-the-loop" to "human-on-the-loop" oversight' as AI grows more capable, while still requiring clear accountability and continuous monitoring.
• Governance basics: clear ownership of each AI use case and an up-to-date inventory of approved AI applications.
• Evidence base: drawn from the second GenA.I. Sandbox cohort, where more than 60% of use cases involved AI agents and half of those were multi-agent; released alongside an insight report sharing Sandbox good practices and lessons.
• The HKMA says the guidance aligns with the Financial Stability Board's AI sound practices.
• The HKMA is exploring a centralised supervisory registry of AI agents and plans a forthcoming discussion paper on a supervisory framework for agentic-AI governance (intent-anchored accountability, a registry of material AI agents, intervention mechanisms) — not yet published.
• Announced alongside a planned AI training-data-sharing platform built with Cyberport and the University of Hong Kong, and the Finance + A.I. Forum co-hosted with Cyberport on 15–16 October 2026.
• Several press reports describe the guidance as replacing the HKMA's 1 November 2019 High-level Principles on Artificial Intelligence with nine principles across the three ART pillars; this tracker could not confirm that specific framing from an HKMA primary source and does not assert it — see hkma-high-level-principles-ai-2019, which remains listed as current pending confirmation.

What did HKMA "ART" Framework / AI supervisory guidance change for banks?

This is the HKMA's first dedicated response to agentic AI findings from its own sandbox programme, moving from the 2019 principles' human-in-the-loop baseline toward oversight calibrated to AI agents' autonomy, and toward infrastructure (a possible agent registry, kill-switch expectations) built for a world where AI acts rather than only recommends.