HKMA ART Framework Puts Agentic AI Under Banking Rules

On October 15, 2026, the Hong Kong Monetary Authority (HKMA) unveiled the ART framework — Accountability, Resiliency and Trust — a supervisory package for artificial intelligence in banking that places agentic AI under explicit governance at one of the world's major financial regulators. According to the HKMA's October 15 announcement, the HKMA ART Framework sits under the regulator's Fintech 2030 strategy and combines new supervisory guidance with a shared AI training-data platform and deeper testing of agentic systems, alongside a forthcoming discussion paper on intent-anchored accountability, a registry of material AI agents, and effective intervention mechanisms.

The timing reflects how fast the technology has moved inside banks. According to the HKMA, more than 60 percent of use cases in the second cohort of its Generative AI Sandbox explored agentic AI — systems that plan and carry out multi-step tasks within defined boundaries and controls. The authority marked the cohort's completion by co-hosting the two-day Finance + AI Forum with Cyberport on October 15 and 16. The HKMA ART Framework is the regulatory answer to what that sandbox revealed: the industry is no longer experimenting with chatbots, but with autonomous systems that touch real workflows.

What the HKMA ART Framework requires

The package has four working parts, according to reporting by PANews. First, the HKMA issued its latest Supervisory Guidance on the Responsible Use of Artificial Intelligence, incorporating insights from the sandbox's second phase. Second, it is building an AI training-data sharing platform with Hong Kong Cyberport and the University of Hong Kong, with data quality assurance as a core design principle. Third, agentic AI applications face further testing through the GenAI Sandbox++ program. Fourth, an upcoming discussion paper will propose a supervisory framework for agentic AI governance, exploring intent-anchored accountability, a registry of material AI agents, and intervention mechanisms.

The guidance itself is organized around nine principles spanning the three ART pillars, according to The Global Banker's coverage. Accountability asks who owns an agent's mandate and actions. Resiliency asks whether the system can fail safely and recover predictably. Trust asks whether operators, customers and supervisors can understand the controls and the evidence behind them. As an explainer of the HKMA ART Framework notes, the announcement moves the conversation beyond model accuracy to the harder problem: what happens when an AI system can use tools, reach data, and participate in a financial workflow.

A registry of material AI agents, and the power to stop them

The most consequential element of the HKMA ART Framework may be the proposed registry of material AI agents. Regulators have spent years debating model risk; a registry of deployed agents — each with a recorded mandate, an accountable owner, and defined intervention points — would be a different instrument entirely, closer to licensing than to auditing. The forthcoming discussion paper will explore exactly how intent-anchored accountability arrangements and effective intervention mechanisms should work under the HKMA ART Framework, according to the HKMA.

Concrete safeguards are already visible in the sandbox trials. According to AsiaDaily's reporting, Standard Chartered subjects AI-generated analysis to professional review, while Nanyang Commercial Bank requires customer authentication before a transaction is finalized. The proposed intervention arrangements add another layer: banks could limit or stop agents and switch to human operators or backup processes when automated activity goes off course. Trial areas already span accessibility, fraud detection, customer service, investment support, loan repayment and risk monitoring.

Why agent builders should pay attention

The HKMA ART Framework lands one week after India's CERT-In warned that AI agents had exceeded their assigned tasks and accessed live business systems without authorization, with attackers weaponizing open-source agent frameworks — a story covered here. The two announcements read as a pair: one regulator documenting what goes wrong when agents operate without guardrails, another building the governance to keep them inside boundaries.

For agent developers, the signal is that regulators are shifting from general statements about responsible AI to operational governance — documenting which agents exist, who is accountable for each, and how humans can intervene. Deputy Chief Executive Arthur Yuen said the progression from generative to agentic AI represents a new stage of banking innovation, and that pairing innovation with clear accountability, resilient operating models and human oversight lets banks realize AI's benefits while preserving trust and stability in the financial system, according to the announcement. Agent identity infrastructure — including protocols for verifiable agent identity covered here — is becoming a compliance question, not just an engineering one.

The discussion paper will be the document to watch: if Hong Kong formalizes agent registries and intervention powers, other financial centers are likely to follow, and the HKMA ART Framework could become the template for how agentic AI gets licensed to operate inside banking.