Hong Kong's central bank writes banking's first rulebook for AI agents: the ART Framework is here
Hong Kong's de facto central bank just gave AI agents a rulebook. The Hong Kong Monetary Authority on Thursday unveiled its “ART” Framework — Accountability, Resiliency, Trust — a set of guardrails for banks deploying AI, with autonomous agents squarely in its sights.
What launched
The announcement, published by the HKMA on Thursday (issued at 14:30 HKT), sits under the authority's Fintech 2030 strategy. Three workstreams make up the ART Framework. First, updated supervisory guidance on the responsible use of AI, drawing on insights from the second cohort of the HKMA's Generative AI Sandbox. Second, an AI training-data sharing platform built with Hong Kong Cyberport and the University of Hong Kong, with data quality assurance as a core design principle. Third, deeper testing of agentic AI applications through the GenA.I. Sandbox++ — and, notably, an upcoming discussion paper on a proposed supervisory framework specifically for agentic AI governance.
Deputy Chief Executive Arthur Yuen called the move from generative to agentic AI “an important new stage of innovation for the banking sector,” and said the framework is meant to give banks “a clear foundation” to develop and deploy AI responsibly, paired with “clear accountability, resilient operating models and effective human oversight.”
The sharpest part: a registry of AI agents
The discussion paper is where this gets interesting. The HKMA says it will explore intent-anchored accountability, a registry of material AI agents, and effective intervention mechanisms. Read plainly: banks may soon have to register which agents are operating inside their systems, pin accountability to what each agent is meant to do, and maintain a working kill-switch. That is agent-level oversight machinery — a level of specificity most AI regulation has so far avoided, preferring to talk about models and data.
It's a meaningful shift. Regulators spent two years writing principles for generative AI chatbots and copilots. The ART Framework is one of the first formal frameworks from a major financial regulator aimed at agents that plan and execute multi-step tasks on their own.
Why now: the sandbox numbers
The timing isn't accidental. The second cohort of the GenAI Sandbox just completed its technical trials, and according to the HKMA, more than 60% of the use cases it explored involved agentic AI — systems doing automated planning and execution of multi-step tasks “within defined boundaries and controls.” The HKMA is showcasing the results at a two-day Finance + A.I. Forum co-hosted with Cyberport on October 15–16. The message: the industry is already moving from chatbots to agents, and the regulator intends to keep pace rather than play catch-up.
Why it matters
Watch this as a jurisdictional signal. Hong Kong is positioning itself as a place where banks can deploy advanced AI with regulatory clarity — guidance, shared data infrastructure, sandbox testing, and soon, agent-specific rules. The ART Framework doesn't ban anything; it builds the tracks. And in a year when regulators on both sides of the Atlantic are still arguing about whether existing rules even cover autonomous agents, Hong Kong just answered with a framework that assumes the question is settled: agents are here, they need registries, and someone has to be accountable for what they intend to do.
Sources: HKMA press release: HKMA announces "ART" Framework to promote responsible A.I. adoption (Oct 15); HK Businesswire: HKMA unveils framework for responsible AI in banking (Oct 15).