Regulation•

Hong Kong's Central Bank Presses Lenders to Deploy AI Agents That Tie Blockchain Transfers to Real Customers

The Hong Kong Monetary Authority is accelerating AI adoption at banks to trace illicit crypto flows, while specialists insist that freezing accounts and filing reports must remain human decisions.

Written By :MINRK
Hong Kong's Central Bank Presses Lenders to Deploy AI Agents That Tie Blockchain Transfers to Real Customers

Photo by panpan ma on Unsplash

1. Hong Kong crypto crime moves up the banking regulator's agenda

The Hong Kong Monetary Authority (HKMA) said on Tuesday, November 17, 2026, that it is speeding up efforts to get banks using artificial intelligence against financial crime, including AI agents that could match blockchain transactions to customer records. The Hong Kong crypto push comes as illicit addresses received at least $154 billion in 2025.

Raymond Chan, the HKMA's executive director for enforcement and anti-money laundering, set out the regulator's thinking in remarks that day. He said the authority is working to quicken the pace at which lenders bring AI into their defences, and that part of that work involves studying how autonomous software agents might connect activity recorded on public blockchains with the information banks already hold about the people and companies they serve.

The initiative reflects a practical problem for the city's banking sector. Stablecoins and other digital assets have made suspicious transfers harder to follow through the financial system, and the tools that compliance departments have relied on for conventional payments were not designed for money that can leave a bank account, cross into a token and then scatter across many wallets within a short period. The regulator's answer is to push institutions toward technology that can keep up with that speed.

2. Why is crypto crime hard for banks to trace?

Banks can follow a transaction across a blockchain, but they often cannot tell who controls the wallet at either end of it. That gap between a visible transfer and an unknown owner is the central obstacle, and it explains why the HKMA is focusing on tools that link on-chain activity to verified identities.

Public blockchains record every movement of funds in a ledger that anyone can inspect. In that narrow sense, crypto is more transparent than cash. What the ledger does not record is a name. An address is a string of characters, and a single person or organisation can create as many of them as it likes. A compliance officer can therefore watch money move from one address to the next without learning anything about the party behind it.

Chan described this as the heart of the matter. "The core challenge obviously lies in linking wallet activities to real-world identities," he said. He added that the difficulty is not simply making that connection but making it quickly enough, before the funds in question have been dispersed. Once money has been split among many wallets or converted into other assets, the chances of stopping or recovering it fall sharply, so the value of an identification depends heavily on when it arrives.

To illustrate the scale of the problem, Chan pointed to research from the blockchain analytics firm Chainalysis. That work found that illicit crypto addresses received at least $154 billion during 2025, an increase of 162% from the previous year. Much of that growth came from activity involving sanctioned entities. Stablecoins, which are tokens designed to hold a steady value against a currency such as the US dollar, accounted for 84% of the illicit total.

The same research offers some perspective on those figures. Chainalysis estimates that illicit activity represented less than 1% of the crypto transaction volume it measured. The absolute sums are large and rising quickly, but they remain a small share of overall activity, which is part of what makes detection hard: the suspicious transfers are a thin stream inside a very large flow of ordinary ones.

3. What the regulator expects AI agents to do

Software agents differ from earlier compliance tools in that they can carry out several tasks in sequence with only limited human supervision. Rather than scoring a single transaction against a fixed rule, an agent can collect information from different systems, compare it and decide what to look at next. The HKMA sees that capacity as suited to the layered and fast-moving nature of crypto-related laundering.

Chan gave an example of the kind of behaviour such systems might catch. Agents could flag patterns such as rapid switching between cryptocurrency and conventional money, a technique that can be used to obscure where funds originally came from. Each conversion on its own may look unremarkable, and the signal only appears when the steps are viewed together and in order, which is the type of analysis that is slow for a human investigator and well suited to automation.

He also said that tools intended to establish who controls a given wallet are being piloted. These address the identification gap directly, aiming to give banks a basis for connecting an on-chain address to a known customer or counterparty rather than leaving it as an anonymous endpoint.

The regulator is not leaving banks to work this out alone. The HKMA has offered technical support and has run workshops for the industry. Its role in this area is that of a supervisor encouraging adoption and helping institutions build capability, with the banks themselves responsible for selecting, testing and operating the systems.

4. Survey shows nine in ten institutions adopting or planning generative AI

Fresh evidence on how far the industry has moved arrived the same day in a report from the Hong Kong Institute for Monetary and Financial Research. The study, which is backed by the HKMA, drew on a survey carried out between January and March that covered 39 financial institutions and eight technology providers.

According to the findings, 90% of the financial institutions polled had either adopted generative AI or planned to do so. Agents are at an earlier stage. Among the business uses of AI that respondents described, 13% involved AI agents that had already been implemented, while 56% involved agents that institutions planned to introduce. The figures suggest that agent-based systems are moving from experiment toward deployment, although most of that deployment still lies ahead.

The report includes a case study relevant to the HKMA's priorities. It describes one bank that used generative AI to investigate possible money laundering by examining transaction records, images and customers' digital activity together. The bank reported that its fraud investigations became faster and more accurate as a result, though it did not provide details to support that assessment.

Enthusiasm was accompanied by reservations. The institutions surveyed raised concerns about security and about their ability to explain decisions made by AI. Explainability carries particular weight in compliance work. When a bank declines a payment or reports a customer to the authorities, it needs to be able to say why, both to the customer and to its supervisor. A system that reaches a conclusion without a clear account of its reasoning is difficult to defend in either setting.

5. Specialists want account freezes left to bank staff

People who work on these systems argue that the technology should be introduced gradually and kept away from the most consequential decisions. Abdul Rafay Gadit, co-founder of ZIG Finance, said banks should test AI agents against past fraud cases and run them alongside their existing monitoring before allowing them to act independently. Replaying known cases shows whether an agent would have caught what the bank already knows was wrong, and parallel operation reveals how its judgments compare with those of the current controls.

Gadit also cautioned against treating a drop in alerts as proof of success. Compliance teams are often overwhelmed by warnings that turn out to be harmless, and a system that produces fewer of them is attractive. He warned that fewer alerts only help if suspicious activity is still being caught. A quieter dashboard that misses real cases would leave a bank worse off, not better.

On the division of labour, Gadit drew a clear line. AI agents could gather transaction records, check names against sanctions lists and prepare cases for investigators. Decisions to freeze an account, reject a payment or report suspicious activity should remain with bank staff. The reasoning is that these actions have direct effects on customers and their money, and the tools must be able to detect suspicious transfers without putting customer funds at risk.

Dominik Clemente, a research engineer at the World Foundation, raised a related point about the humans in that loop. He said banks should verify that the person approving a flagged transaction is who they claim to be, particularly when approvals are given remotely. Using the example of a video call, he said he would want to know that the approver is really the person appearing on camera. A human checkpoint offers little protection if the identity of the human cannot be trusted.

6. A Federal Reserve governor's warning about autonomous agents

Concerns about agents acting on their own are not confined to Hong Kong. On Monday, November 16, Federal Reserve Governor Michael Barr warned that AI agents could speed up bank runs by moving deposits without human approval. A bank run occurs when many depositors withdraw funds at once, and automation that shifts money on a customer's behalf could shorten the time in which that happens.

Barr said banks had already seen agents attempt to act on accounts in circumstances where the institution could not verify that the customer had consented. His remarks concern a different risk from money laundering, but they point to the same underlying questions that Hong Kong's regulator is now examining: how a bank can know that an agent is authorised to do what it is attempting, and who answers for the outcome when it is not.

7. What comes next for Hong Kong's AI plans

Those questions are the subject of Project SAGE, which the HKMA launched on November 2. The project studies how financial AI agents should prove their identity and their authority to act, and who is responsible when things go wrong. Its findings are likely to shape how far banks in the city are permitted to let agents operate without a person approving each step.

For now, the direction from the regulator is clear even where the details are still being worked out. The HKMA wants lenders to adopt AI more quickly, it is examining agents that can connect blockchain transactions with customer records, and it is supporting pilots of tools that identify who controls a wallet. The survey data indicates that most institutions intend to follow, with agent deployments planned by a majority of respondents.

The remaining tension is between speed and control. Chan's argument is that identifications must happen fast enough to stop funds before they scatter, which favours automation. The specialists' argument is that actions affecting customers must stay with accountable staff, which limits how much can be automated. How banks reconcile those two demands, and how Project SAGE defines responsibility for agents, will determine what the next generation of crypto crime controls in Hong Kong looks like.

Related Articles

NEWSLETTERS

Don't miss another story.

Subscribe to the MINRK Newsletter today.

Tick the newsletters you want. No selection? You'll get MINRK Headlines.

By signing up, you will receive emails about MINRK products and you agree to our terms of use and privacy policy.

MINRK Daybook

Market analysis for crypto traders and investors.

EVERY WEEKDAY

MINRK for Advisors

Defining crypto, digital assets and the future of finance for financial advisors.

EVERY THURSDAY

MINRK Protocol

Exploring the tech behind crypto one block at a time.

WEEKLY

MINRK Long & Short

A must read for institutions. Insights, news and analysis delivered weekly.

EVERY WEDNESDAY

MINRK Headlines

The biggest crypto news and ideas of the day.

EVERY WEEKDAY

MINRK Policy

Examining the intersection of cryptocurrency and government.

WEEKLY

MINRK Research

Join thousands of readers who rely on MINRK for data-driven insights on the latest digital asset trends.

MONTHLY