Hong Kong Sets Out New AI Governance Framework for Banks as Agentic Systems Expand
The HKMA framework addresses governance, data quality and operational risks as banks test agentic AI for customer service, fraud detection and other financial activities.
Hong Kong's banking regulator has introduced a framework intended to support responsible artificial intelligence adoption, as financial institutions explore systems capable of planning and carrying out multistep tasks with less direct human intervention.
The Hong Kong Monetary Authority (HKMA) announced its "ART" framework on October 15, outlining three central principles: accountability, resilience and trust. The initiative is designed to help banks use AI while maintaining appropriate governance and operational safeguards, according to the regulator's official announcement.
The framework comes as banks increasingly move beyond conventional chatbots and generative AI tools toward what is commonly called agentic AI. Such systems can be designed to plan tasks, access approved tools and data, and perform a sequence of actions within defined boundaries. Potential applications include customer service, risk analysis, fraud detection and internal administrative work.
The HKMA said more than 60% of the use cases explored in the second cohort of its Generative AI Sandbox involved agentic AI. The sandbox allows participating institutions to test applications and assess their benefits and risks before wider deployment.
The regulator's approach is built around three principles. Accountability concerns who is responsible for an AI system's design, decisions and actions. Resilience focuses on the ability of banks to manage failures, unexpected behavior and operational disruptions. Trust involves the quality of data, the reliability of outputs and appropriate oversight of systems used in sensitive financial activities.
These considerations become more important as AI tools gain access to a wider range of data sources and software functions. A system that merely generates a text response presents different operational risks from one that can initiate workflows, retrieve customer information or make recommendations that influence financial decisions.
The HKMA said it had issued updated supervisory guidance on responsible AI adoption. It is also working with Hong Kong Cyberport and the University of Hong Kong on an AI training-data sharing platform, with data quality assurance identified as a central design principle. The regulator plans further testing through its GenAI Sandbox++ and intends to publish a discussion paper on a proposed governance framework for agentic AI.
One issue is how to preserve clear responsibility when an AI agent performs several connected actions. Banks may need to define the agent's purpose, restrict the tools and information it can access, establish intervention mechanisms and ensure that human staff can review or stop activity when necessary.
Data quality is another concern. AI systems can produce unreliable results if their training data are incomplete, outdated or poorly suited to a task. In banking, errors may affect customer onboarding, compliance checks, fraud investigations or risk assessments. Strong testing and monitoring are therefore essential before institutions rely on AI for consequential decisions.
The HKMA's work also reflects the broader challenge of balancing innovation with financial stability. Banks are under pressure to improve efficiency and customer experience, but they must also protect confidential information, comply with financial regulations and maintain service continuity. AI adoption that reduces costs but creates poorly understood risks could undermine those objectives.
The authority's October 16 remarks at the Finance + AI Forum reported examples of potential benefits from sandbox trials. These included a reduction in customer-onboarding time at one bank, positive feedback on a wealth-management chatbot and shorter analysis cycles for a market-risk agent. Such examples indicate possible efficiency gains, but outcomes from trials do not automatically demonstrate that similar results will occur across the industry.
Hong Kong is seeking to strengthen its role as an international financial center while developing its technology and digital-finance ecosystem. Clearer AI governance may help banks experiment with new tools while providing a more consistent basis for assessing risks.
The framework does not mean that all AI applications will be approved for unrestricted use. Individual banks remain responsible for ensuring that their systems comply with relevant requirements and that appropriate controls are in place.
The next challenge will be translating high-level principles into practical oversight, especially as AI agents become more capable and interconnected. For Hong Kong's financial sector, the ability to demonstrate accountability and reliability may prove as important as the speed of technological adoption.