HKMA issues new AI supervisory guidance under 'ART' framework as banks test agentic AI

Implications for lenders

Lenders that use AI in underwriting, collections or customer contact may need clearer records of who is accountable for each model or agent and where staff can step in, since the HKMA plans a registry of material AI agents. With more than 60% of sandbox use cases already agentic, firms partnering with banks may face more questions on AI governance in due diligence and outsourcing reviews.

Implications

Facts

The Hong Kong Monetary Authority (HKMA) announced the 'ART' framework on 15 October 2026 to support responsible adoption of artificial intelligence in the banking sector, built on accountability, resilience and trust. It has three parts. First, the HKMA issued its latest Supervisory Guidance for Responsible Adoption of Artificial Intelligence, drawing on the second cohort of its Generative AI (GenAI) Sandbox, whose technical trials the HKMA says were completed with positive results. Second, an AI training data sharing platform is being built with Cyberport and the University of Hong Kong, with data quality assurance as a core design principle. Third, the HKMA will continue testing agentic AI through a 'GenA.I. Sandbox++' and will publish a discussion paper on a proposed supervisory framework for agentic AI governance, covering intent-anchored accountability, a registry of material AI agents and intervention mechanisms (source: HKMA).

According to the HKMA, more than 60% of the use cases in the second sandbox cohort explored agentic AI. ¹ The release gives no date for the discussion paper and no consultation period. Deputy Chief Executive Arthur Yuen called the move from generative to agentic AI 'an important new stage of innovation' for banks.

Analysis

Across the two financial centres, the approaches arrived a week apart but differ in form. The Monetary Authority of Singapore issued final Guidelines on AI Risk Management on 7 October 2026 for all financial institutions, with oversight and risk-management expectations applying from 7 October 2027 and life-cycle controls from 7 October 2028, and plans to consult on agentic AI in 2027. The HKMA's package is addressed to banks, combines guidance with shared infrastructure (a training-data platform) and sandbox testing, and moves straight to agentic AI, where a registry of material AI agents would be a new supervisory tool. With more than 60% of sandbox use cases already agentic, Hong Kong supervisors are setting expectations while the technology is still in trials rather than after wide deployment.

Counterpoints and uncertainties

The release summarises the supervisory guidance without setting compliance dates, so the practical burden depends on the guidance text and on the later discussion paper, whose timing is not given. The framework is addressed to banks; licensed money lenders, which operate under the Money Lenders Ordinance rather than HKMA supervision, are outside its direct scope, so AI practices may diverge across Hong Kong's consumer-lending market. Sandbox results may also not carry over to production systems at scale.