HKMA AI transaction monitoring guidance is often described as if one circular had made artificial intelligence a supervisory expectation. The record is more nuanced. Over three years the Hong Kong Monetary Authority has issued a thematic review, a request for feasibility studies, a progress circular and, most recently, a June 2026 report and workshop on agentic AI. Each step encourages adoption and sets conditions, but none says every bank must deploy AI.
This article sets out the dated sequence, what banks have actually been asked to do, and the controls the HKMA expects around AI used in monitoring.
The sequence of HKMA documents
On 17 April 2024 the HKMA shared a thematic review of transaction monitoring systems, covering governance, data quality, detection scenarios, threshold setting and periodic review, with a section on how banks use AI to optimise performance. On 9 September 2024 it asked banks with significant Hong Kong operations to carry out a feasibility study on using AI in suspicious activity monitoring, with the study and implementation plan signed off at board level.
A circular of 19 November 2025 reported that 48 authorized institutions had completed the feasibility work. It noted that some banks had replaced reliance on rules-based monitoring with a more holistic approach, while others were adding AI in phases, mainly for risk detection and alert prioritisation. The HKMA announced workshops on these approaches and on using generative AI to compile suspicious transaction reports. On 22 June 2026 it issued a circular and report, Supporting AI Adoption in Fighting Financial Crime, and announced a second workshop on agentic AI use cases, held on 23 June 2026.
What the HKMA is and is not asking
Read closely, the language is one of support. The HKMA asked for feasibility studies, shared good practice, and said it would continue tracking adoption and the effectiveness of banks' monitoring systems. It has not mandated a technology. What it has made clear is that traditional approaches to monitoring face pressure from faster payments, mule account networks and synthetic identities, and that banks should be evaluating AI as part of end-to-end transformation of their AML/CFT operating model.
In June 2026 the HKMA also said new use cases being explored should be reported to it when called for. Banks should therefore keep a current record of what they are testing and why, because supervisory contact is likely to ask.
The controls expected around AI
The HKMA's thematic review of sanctions screening, published on 16 March 2026, gives the clearest statement of conditions. As summarised by PwC China, it observed growing use of AI to improve screening efficiency, including automated false-positive closure, provided the logic is explainable and supported by adequate quality assurance. The same logic can reasonably be expected to apply to monitoring.
In practice, banks should be able to show model governance, documented rationale for how alerts are scored or closed, human oversight at defined points, testing against known cases, and ongoing performance monitoring. A bank that cannot explain why an alert was deprioritised will struggle to defend that decision to a supervisor, however accurate the model.
A practical agenda for banks
Banks that completed feasibility studies should check that their implementation plans still match HKMA's direction, and that board reporting covers progress and risks. Those with rules-based systems should document how thresholds were set and tuned, since the thematic review concentrates on those basics, and AI cannot rescue weak data or poorly scoped scenarios.
Whichever route a bank takes, the foundations are the same: clean data, calibrated detection, documented investigation and a clear audit trail. The June 2026 circular's invitation to explore agentic AI should be read as an invitation to experiment under governance, not as a deadline. Seen as a whole, HKMA AI transaction monitoring policy rewards banks that can govern and explain their models.
