HKMA introduces ART framework for responsible AI in banking

The ART framework names three principles: Accountability, Resilience and Trust.[2] Accountability means clear ownership, governance, decision-making authority and assurance throughout an AI system’s lifecycle.[2] Resilience calls for AI that is secure, robust and well controlled, operates safely, fails safely and recovers effectively.[2] Trust concerns fair, transparent, explainable and privacy-conscious AI use that responds to customers and other affected parties.[2] The HKMA says the framework is intended to support responsible AI adoption in Hong Kong’s banking sector.[1]

The guidance says a bank’s board should approve an AI strategy that states the bank’s approach and risk appetite for AI adoption.[2] Banks should assign an owner to each deployed AI use case and provide effective human oversight of AI-generated outcomes.[2] The HKMA says reviewers need the competence, information, time and authority to challenge, override or stop an AI system.[2] Banks should also keep an up-to-date inventory of AI use cases approved for deployment, restrict use of solutions absent from it and implement controls to manage shadow AI adoption.[2] The HKMA says banks should update their AI policies, procedures and controls in line with the guidance in a timely manner.[2]

The new guidance supersedes the HKMA’s 2019 high-level AI principles and incorporates what the HKMA describes as the latest guidance issued by international bodies.[2] The HKMA expects authorized institutions to apply the ART principles proportionately to their AI adoption and manage the associated risks.[2] That means the guidance sets a common framework while asking institutions to account for their own AI adoption and risk management.[2] For bank teams, the practical work described in the guidance includes setting strategy and risk appetite, assigning use-case ownership, maintaining an approved-use inventory and ensuring that human reviewers can intervene.[2]

HKMA Deputy Chief Executive Arthur Yuen said the shift from generative AI to agentic AI is an important new stage of banking innovation.[3] He said pairing innovation with accountability, resilient operating models and human oversight can help banks realize AI’s benefits while maintaining trust and financial-system stability.[3] The HKMA plans further testing of agentic AI applications through a sandbox programme and an AI training-data sharing platform in collaboration with Cyberport and the University of Hong Kong.[3] The HKMA said more than 60 percent of use cases explored agentic AI applications in the second cohort of its generative AI sandbox trial.[3] It is showcasing the sandbox’s positive results during a two-day Finance+AI Forum co-hosted with Cyberport.[3]