Hong Kong's banking regulator is accelerating its push for banks to use artificial intelligence against financial crime, including suspicious transfers involving stablecoins and other digital assets. The effort brings together two challenges for lenders: detecting potentially illicit movements quickly and ensuring that automated systems do not put customers' money at risk. The work described so far is under examination, rather than a completed monitoring system or a new mandate for banks.
On November 17, Raymond Chan, the Hong Kong Monetary Authority's executive director for enforcement and anti-money laundering, said the regulator was examining how AI agents could combine blockchain transactions with customer records to identify possible crime. The aim is not simply to observe that digital assets have moved. It is to connect that activity to people quickly enough to intervene before the funds are dissipated, according to Chan.
From blockchain trails to real-world identities
That distinction matters because following activity on a blockchain is not the same as establishing who controls a wallet. Customer information could help bridge that gap, but timing is central to the problem Chan described. He also said identity tools intended to help determine wallet control were being piloted. Those pilots are separate from a claim that the regulator has already established an AI system capable of identifying every suspicious transfer.
One pattern AI agents could flag, Chan said, is rapid conversion between cryptocurrency and conventional money that might conceal the origin of funds. Such a pattern would be a reason to investigate, not proof on its own that a crime had occurred. The proposed use of customer records alongside transaction data is directed at making that investigation more informative than a blockchain trail alone.
Chan cited Chainalysis estimates to describe the scale of the problem. Illicit crypto addresses received at least $154 billion in 2025, according to the estimates, an increase of 162% from 2024. Stablecoins accounted for 84% of the total, with activity involving sanctioned entities a major driver. Chainalysis also put illicit activity at less than 1% of the transaction volume it measured, placing the dollar figure alongside a narrower share of measured activity.
Banks are adopting AI beyond crypto monitoring
The anti-crime effort sits within a broader push to help banks use AI. A Hong Kong Institute for Monetary and Financial Research report released November 17 said the HKMA had provided technical support and workshops for deployment. The measures were not described as confined to cryptocurrency monitoring. That broader scope is important: a bank's experience with AI in another area does not establish that a digital-asset monitoring tool has been completed.
The institute described one bank using generative AI in possible money-laundering investigations, drawing on transaction records, images and customers' digital activity. The bank reported that its fraud investigations became faster and more accurate, although the report did not quantify those improvements. The example shows an existing investigative use of AI, while leaving open how much the technology improved results in that case.
The report drew on a January-to-March survey of 39 financial institutions and eight technology providers. Among the financial institutions, 90% had adopted generative AI or planned to adopt it, a figure that includes future plans as well as existing use. Across business use cases, 13% had implemented AI agents and 56% planned to do so. Banks also raised concerns about security and how to explain decisions produced by AI.
Testing detection without surrendering control
ZIG Finance co-founder Abdul Rafay Gadit recommended testing agents against historical fraud cases and running them alongside existing monitoring systems before permitting independent action. He suggested comparing what the systems detect, what they miss and how long investigations take. That approach would give banks more than an alert count to evaluate: a reduction in alerts alone would not establish that a system was catching the right cases.
In Gadit's view, agents could gather transaction records, check sanctions lists and prepare files for investigators. Bank employees, however, should retain decisions on freezing accounts, rejecting payments and reporting suspicious activity. His recommendation separates investigative preparation from decisions that can directly affect customers or trigger a report to authorities. It is a proposed safeguard from an industry participant, not a new HKMA rule described in the report.
World Foundation research engineer Dominik Clemente identified another risk at the approval stage: banks need to establish that the person authorizing a flagged transaction is genuinely the claimed approver, particularly when sign-off is remote. Separately, the HKMA launched Project SAGE on November 2 to examine the identity and authority of financial AI agents and accountability when problems arise. The earlier project addresses related governance questions, rather than constituting the November 17 anti-crime initiative.
