AiStaffo

13% of Hong Kong Financial Firms Run Agentic AI, 56% Plan To

Stock photo: office desk documents laptop

The Hong Kong Institute for Monetary and Financial Research (HKIMR), the research arm of the Hong Kong Academy of Finance, released a report on 17 November 2026 on AI use in the city's financial services. It found that 13% of surveyed institutions had already implemented agentic AI, meaning software that can carry out multistep tasks with limited human input, and that 56% planned to. Among surveyed institutions, 86% had adopted or planned to adopt generative AI tools for regulatory compliance. The sample is small and the figures describe surveyed firms, not the whole sector. For businesses that automate routine work, the report signals that autonomous tools are moving from pilots into daily operations, and that regulators expect named owners, testing and the ability to intervene.

In short

  • HKIMR's report, released 17 November 2026, found 13% of surveyed Hong Kong financial institutions had implemented agentic AI and 56% planned to.
  • 86% of surveyed institutions had adopted or planned to adopt generative AI tools for regulatory compliance.
  • The SFC's separate survey of 20 large firms found 30% already use agentic AI, so the two samples should not be compared directly.
  • The HKMA's agentic AI discussion paper proposes testing, clear accountability and the ability to intervene, and it is still a consultation.
  • Any business that automates work should name an owner, log automated actions and write down what each tool may change.

The report is titled From Assistance to Autonomy: A.I. Adoption, Governance, and Compliance in Hong Kong's Financial Services Industry. According to The Standard, HKIMR based it on a survey and interviews carried out between January and July 2026 with financial institutions, technology firms and regulatory technology providers. The study is the latest in a series HKIMR has run on the topic, and the full text is published on the websites of the Academy of Finance and HKIMR.

What the survey found

The headline numbers are about adoption. According to SCMP, 90% of surveyed institutions said they had adopted or were planning to adopt generative AI, while The Standard reported the same figure as up from 75% in 2025. Generative AI is the kind of tool that drafts text, summarises documents and answers questions. Agentic AI goes further, because it pursues a goal and works through steps on its own.

On compliance, the picture is similar. According to The Standard, 86% of surveyed institutions had adopted or planned to adopt generative AI tools for regulatory compliance, and more than 70% said they were interested in exploring agentic AI for compliance over the next three to five years. OpenGov Asia noted that more than half of the institutions said they would prioritise generative AI while also exploring agentic AI.

Two surveys, two different pictures

A second Hong Kong survey published in the same month gives a higher figure for agentic AI. It comes from the Securities and Futures Commission (SFC), which surveyed 20 large financial firms with 78 affiliated licensed corporations between July and August 2026. The table below sets the two side by side, because they measure different groups.

ItemHKIMR report (released 17 November 2026)SFC large firm survey (July to August 2026)
SampleFinancial institutions, technology firms and regtech providers, surveyed and interviewed January to July 202620 large financial firms with 78 affiliated licensed corporations
Agentic AI in use13% already implemented30% already use it in business operations
Agentic AI planned or piloting56% plan to introduce it25% are piloting it
Expected accelerationNot reported in the sources reviewed95% expect AI use to speed up over the next 12 months

The gap is not a contradiction. The SFC sample is made up of large firms, and the HKIMR sample is wider but smaller. Reports also differ on the HKIMR sample size. SCMP wrote that the report surveyed 47 financial institutions and drew on 19 interviews. The breakdown given for this article, 39 financial-sector respondents and eight service providers, is not confirmed in the sources reviewed. Readers should check the full HKIMR report before quoting the sample.

The regulatory backdrop

Regulation moved in the same weeks. On 15 October 2026 the Hong Kong Monetary Authority (HKMA), the banking regulator, announced its ART Framework, built on accountability, resilience and trust. The HKMA's discussion paper on agentic AI governance, published on 2 November 2026, sets out a preliminary view that all agentic AI deployments should undergo testing. It also raises risks such as unclear accountability, unintended actions, data leakage and failure to contain incidents. The paper is a consultation, not final rules. The HKMA's guidance is addressed to banks. Other financial firms and non-financial businesses should read it as a signal of what supervisors will ask, not as a binding rule for their own operations.

What changes for a business that automates work

The report is about banks, insurers and asset managers, but the practical questions apply to any firm that lets software act on its behalf. Three changes are worth noting.

First, an automation needs a named owner. The HKMA's preliminary view is that AI should stay accountable to people outside the system. For a small firm, that means one person who can explain what a workflow does, what data it touches and who can switch it off.

Second, automated steps need a log and a test before they run live. The discussion paper mentions a record of each AI deployment and the ability to intervene. A business that lets a tool send payment reminders or match invoices should keep a record of what it sent or changed, and test the tool on sample data first.

Third, the scope of what a tool may do should be written down. The HKMA's discussion paper refers to an AI operating within its authorised scope and delegated authority. Most businesses can apply the same idea by listing, for each workflow, what the tool may read, what it may change and what must go to a person.

There are limits. The report does not show that agentic AI saves money or reduces errors for the firms surveyed. Automation is not the right first step where the process is unclear, where the data is poor or where a wrong action would be costly and hard to reverse. In those cases, fixing the process and the data comes first.

What to watch next

Three items will show how the rules develop. The HKMA's discussion paper is open for industry feedback, and the consultation closing date is not stated in the sources reviewed. The HKMA has also launched Project SAGE, which stands for Securing Agentic Finance with Governance and Enablement, along with a cross-sectoral expert group. The HKMA plans to extend agentic AI trials through its GenAI Sandbox++. Business owners who want to follow the topic should check the HKMA and SFC websites for consultation notices and final guidance.

How AiStaffo would automate this

AiStaffo builds AI workers for routine business work such as invoice matching, payment follow-ups, reconciliations and reports. In practice, the first step is to connect the tool to the systems a business already uses, such as its accounting package, inbox or document store, and to set the limits of each task in writing. The AI worker then runs the routine steps automatically and records what it did. A person still approves exceptions, owns each workflow and can pause or switch off any step. AiStaffo offers free tools that help owners and finance teams map these steps before anything runs live. Book a free automation audit.

Questions people ask

What is agentic AI in simple terms?
Agentic AI is software that can work toward a goal through several steps with limited human input. Generative AI mainly produces text, summaries or answers when asked. The HKIMR report treats the two as separate stages of adoption, with agentic AI as the next step.
Does the HKMA rule apply to my business?
The HKMA guidance is addressed to banks in Hong Kong, and the agentic AI paper is a discussion paper, not final rules. Other businesses are not directly covered by it. The practical questions it raises, such as ownership, logs and limits, are still useful for any firm that automates work.
Why do the Hong Kong surveys show different percentages?
The HKIMR survey covered a wider group of firms and providers from January to July 2026. The SFC survey covered 20 large financial firms from July to August 2026. Different samples give different results, so the figures describe each group only.
Is agentic AI worth it for a small business?
The sources do not show a general payoff for businesses of any size. Automation makes most sense where a process is clear, the data is reliable and a wrong action can be reversed. Where the process is unclear or the data is poor, fix those first.
What should a business check before letting AI act on its own?
List the data each tool can read, the actions it can take and the cases that must go to a person. Test the workflow on sample data and keep a log of what it sends or changes. Name one owner who can pause the tool.

Book a free automation audit

Thirty minutes. We look at one process you run every week and tell you exactly what an AI worker would take off your desk, and what it would not.

We use these details only to reply to this request and prepare your audit, and keep them for up to two years. We send no marketing unless you tick the box above. Privacy policy

ai agentsai governanceai regulationcompliance automationhong kong