top of page
fnlogo.png

Hong Kong Study Highlights Growing Interest in Agentic AI for AML Compliance

8 hours ago
3 min read

Hong Kong’s financial services industry is increasingly exploring the use of generative and agentic artificial intelligence (AI) in regulatory compliance, including customer due diligence, transaction monitoring and fraud detection, according to a report published on 17 November 2026 by the Hong Kong Institute for Monetary and Financial Research (HKIMR).


AI for AML

The report, titled *From Assistance to Autonomy: A.I. Adoption, Governance, and Compliance in Hong Kong’s Financial Services Industry*, examines the adoption of AI across the financial sector and considers the governance and operational challenges associated with increasingly autonomous systems.


The research was based on a survey of 47 organisations, comprising 39 financial institutions and eight technology and service providers, together with 19 in-depth interviews conducted between January and July 2026. Among the surveyed institutions that had adopted or planned to adopt generative AI, 86% were implementing or considering its use for compliance-related activities.


Interest in agentic AI was also significant. Over the next three to five years, 78% of respondents expressed interest in using agent-based applications for customer onboarding and due diligence, while 74% identified transaction monitoring and fraud detection as potential areas of application. These figures reflect prospective interest rather than confirmed operational results.


The report highlights the potential for agentic AI to move beyond supporting individual tasks and undertake more complex, multi-step activities. In an AML context, this could include collecting and reviewing customer documentation, identifying inconsistencies, preparing risk assessments and supporting the investigation of suspicious transactions. However, the increasing autonomy of these systems also creates questions concerning accountability, decision-making authority and the reliability of automated assessments.


A central issue identified is the need to ensure that decisions supported by AI remain transparent and capable of being independently reviewed. Financial institutions must be able to distinguish information extracted directly from source documents from conclusions generated by an AI system. This is particularly important where customer records contain conflicting information or where a risk assessment depends on incomplete evidence.


The study also emphasises the importance of maintaining adequate records of automated processes, including the information considered by the system, the assessment it generated and the final decision made by the responsible employee. Such records would enable institutions to reconstruct decisions, investigate errors and assess the impact of changes to AI models.


Hong Kong’s supervisory authorities have already begun addressing these developments. The Hong Kong Monetary Authority (HKMA), in its June 2026 communication on the use of AI in combating financial crime, highlighted initiatives involving dynamic risk assessment, facial watch lists and the detection of corporate mule accounts. Financial institutions with significant operations in Hong Kong had also been required to submit implementation plans, which the HKMA indicated should remain under review as new applications and developments emerge.


In October 2026, the HKMA introduced its Accountability, Resilience and Trust (ART) framework, alongside further work through its GenAI Sandbox++. The authority also opened discussion on governance arrangements specifically addressing agentic AI, including accountability for the actions of autonomous systems, the potential registration of material AI agents and mechanisms allowing institutions to intervene when necessary.


Separately, the Securities and Futures Commission (SFC) has issued guidance for licensed corporations using language models in connection with regulated activities. Its framework addresses senior management responsibility, proportionate risk controls, validation before deployment and following material changes, ongoing monitoring, cybersecurity and reliance on external technology providers.


The international regulatory perspective is also developing. On 15 October 2026, the Financial Stability Board published a report setting out 12 sound practices for the responsible adoption of AI by financial institutions. The recommendations address governance, risk management, documentation, data quality, oversight and the assessment of risks throughout the technology lifecycle.


Cyprus Company Formation

Despite the anticipated benefits, the report identifies challenges that may affect wider adoption, including concerns over explainability and security. Institutions must also consider their dependence on external technology providers and ensure that appropriate arrangements exist if an AI service becomes unavailable or produces unreliable results.


The findings underline that the introduction of agentic AI into AML operations should not be assessed solely by reference to processing speed or cost savings. Financial institutions will also need to evaluate whether automated systems improve the quality of risk assessments, identify relevant discrepancies and support defensible decisions. Human oversight remains important, particularly where an automated recommendation could affect customer relationships, risk classifications or the handling of suspicious activity alerts.


The growing interest in agentic AI therefore represents both an opportunity and a governance challenge for the financial sector. While autonomous systems may increase the efficiency of customer due diligence and financial crime monitoring, their deployment will require clear allocation of responsibility, appropriate access restrictions, reliable audit trails and effective mechanisms for human intervention.

By fLEXI tEAM

Comments


bottom of page