The HKMA has announced its ART Framework and issued new supervisory guidance on responsible AI adoption in banking

Key takeaways

• The ART Framework stands for Accountability, Resilience and Trust, and sits under the HKMA's Fintech 2030 strategy.
• The HKMA says more than 60% of the use cases in the second GenA.I. Sandbox cohort explored agentic A.I.
• A discussion paper on a proposed supervisory framework for agentic A.I. governance is described as upcoming. It is a proposal, not a rule.
• The release states no effective date, no deadline and no sanction.

What did the HKMA announce, and is it binding?

The HKMA announced the ART Framework on 15 October 2026 "to support the responsible adoption of artificial intelligence (A.I.) in the banking sector". It describes the aim as an AI-enabled banking ecosystem with Accountability, Resilience and Trust. The release also says AI systems "are becoming more autonomous, with greater access to data, tools and workflows", and presents the framework as a way to help banks seize AI opportunities while maintaining governance and operational resilience.

The release does not say whether the supervisory guidance binds banks, how the HKMA would supervise against it, or when any expectation takes effect. It describes the document only as the latest Supervisory Guidance for Responsible Adoption of Artificial Intelligence. A bank should treat the legal weight of the guidance as something to confirm in the document itself.

What are the three initiatives under the ART Framework?

The release lists three. First, the HKMA "Issued the latest Supervisory Guidance for Responsible Adoption of Artificial Intelligence", drawing on insights from the second cohort of the Generative A.I. (GenA.I.) Sandbox. Second, an AI training data sharing platform, described as in collaboration with Hong Kong Cyberport Management Company Limited and the University of Hong Kong, with data quality assurance as a core design principle. Third, further testing of agentic AI applications through the GenA.I. Sandbox++, together with an upcoming discussion paper on a proposed supervisory framework for agentic A.I. governance.

The HKMA also co-hosted a two-day "Finance + A.I. Forum" with Cyberport on 15 and 16 October, which it says marked the completion of the technical trials under the second sandbox cohort, with positive results. We cover the Sandbox++ programme separately in our article on GenA.I. Sandbox++.

What does the HKMA say about agentic AI?

The release reports that more than 60% of the use cases explored in the second sandbox cohort involved agentic AI, which it describes as applications such as "automated planning and execution of multi-step tasks within defined boundaries and controls". The HKMA reads this as the industry actively pursuing advanced AI applications.

On supervision, the release announces an upcoming discussion paper on a proposed supervisory framework for agentic AI governance. It says the paper would explore intent-anchored accountability, a registry of material AI agents and effective intervention mechanisms. Those are topics for discussion, not adopted requirements. Mr Arthur Yuen, Deputy Chief Executive of the HKMA, is quoted as saying "By pairing innovation with clear accountability, resilient operating models and effective human oversight, banks can fully realise the benefits of A.I. while maintaining the trust and stability for Hong Kong's financial system."

What are the dates and figures in the release?

What does this mean for banks and their AI vendors?

For banks, our reading is that the release points to themes the HKMA may focus on: accountability for AI outcomes, operational resilience of AI-dependent processes, and human oversight. Because the guidance text is not in the release, the specific expectations on model risk, third-party AI and governance are still to be read from the guidance.

For AI vendors and fintechs serving Hong Kong banks, the practical point is that a bank customer could ask how an agent is bounded, who is accountable for its actions, and how a human can intervene. The proposed registry of material AI agents is only a topic in a forthcoming paper, so vendors cannot yet treat it as a requirement.

What should banks and vendors do now?

• Obtain the Supervisory Guidance for Responsible Adoption of Artificial Intelligence from the HKMA and compare it with your AI governance policy.
• List AI systems that plan or act with some autonomy, and record who is accountable for each and where a human can intervene.
• Watch for the HKMA discussion paper on agentic AI governance and consider responding when it appears.
• If you are a vendor, prepare plain answers on agent boundaries, data quality and human override for bank due diligence.

What we did not verify

What we opened: the HKMA press release of 15 October 2026 on the HKMA website, checked on 25 October 2026.

What we did not open: the Supervisory Guidance for Responsible Adoption of Artificial Intelligence itself, which the press release does not link, the Sandbox cohort reports, the forthcoming discussion paper, and any bank or industry response. We do not know the guidance's length, scope, binding status or effective date.

What we refuse to claim: that the guidance is binding or enforceable, that the HKMA will require an AI agent registry, or that any deadline or sanction applies. The release describes the agentic AI paper as upcoming; we did not see it and make no claim about its publication status.

Related briefings

• AI Regulation News hub
• Asia-Pacific AI regulation tracker
• Finance professionals and AI regulation
• Hong Kong GenA.I. Sandbox++ and four regulators
• HKMA generative AI and consumer protection duties
• PCPD AI compliance check of 60 organisations and agentic AI
• SFC circular on AI-enabled cyberattacks