HKMA Centers Accountability and Trust in AI Governance

The Hong Kong Monetary Authority (HKMA) has introduced the “ART” Framework to guide the responsible adoption of AI in the banking sector, built around three pillars: Accountability, Resilience, and Trust.

A key component of the new guidance, which supersedes the regulator’s 2019 high-level principles, is a heightened focus on governance and accountability. The HKMA expects boards and senior management to remain accountable for their institution's AI adoption by setting effective strategy and governance arrangements. Banks are expected to assign a clear “owner” for each deployed AI use case and maintain an up-to-date inventory of approved applications.

The HKMA notes that as AI capabilities advance, human oversight must evolve. Depending on the risks involved, institutions may move from ‘human-in-the-loop’ toward ‘human-on-the-loop’ or ‘human-in-command’ arrangements. Reviewers involved in such interventions must have sufficient competence, information, time, and authority to challenge, override, or stop the AI system.

The updated guidance reflects the rapid evolution from generative AI to “agentic AI” – systems capable of planning and executing multi-step tasks, using tools, and accessing data with varying degrees of autonomy. Noting that more than 60% of use cases in the second cohort of its Generative AI Sandbox involved AI agents, the HKMA published a Concept Note specifically addressing the governance of agentic AI.

To govern these increasingly autonomous systems, the HKMA is  exploring “intent-anchored accountability.” This concept would require banks to establish a clear connection between an AI agent’s actions and the underlying business intent, ensuring a clear allocation of responsibility throughout the application's lifecycle. The regulator is also exploring the creation of a centralized supervisory registry for material AI agents to ascertain accountability arrangements and facilitate intelligence sharing across the banking sector.

Under the Resilience and Trust pillars, the HKMA expects institutions to implement risk-proportionate management, robust data governance, and third-party risk controls. For material or higher-risk use cases, banks must build intervention mechanisms, such as a “kill switch,” to suspend or decommission errant AI behavior. In addition, AI adoption must be ethical, fair, and transparent, with human reviewers empowered to challenge and correct AI-generated outcomes to ensure customers are treated fairly.

“The progression from generative to agentic AI represents an important new stage of innovation for the banking sector,” said Arthur Yuen, Deputy Chief Executive of the HKMA. “By pairing innovation with clear accountability, resilient operating models and effective human oversight, banks can fully realise the benefits of AI while maintaining the trust and stability for Hong Kong's financial system.”