The Hong Kong Monetary Authority and HKUST Business School have examined how agentic AI could improve financial-crime detection, source-of-wealth checks and suspicious transaction reporting.
The Hong Kong Monetary Authority (HKMA) and HKUST Business School have jointly hosted a workshop on the application of agentic artificial intelligence in anti-money laundering and counter-financing of terrorism.
More than 200 participants attended, including representatives from licensed banks, industry professionals, academics and students. Speakers included representatives from Bank of China (Hong Kong), Deutsche Bank, UBS and Ant Bank.
The workshop focused on moving AI beyond basic efficiency improvements towards predictive and preventive risk detection.
Moving beyond rules-based monitoring
Conventional transaction-monitoring systems rely heavily on predetermined rules and thresholds. Although these systems remain important, they can produce large volumes of false positives while failing to detect more complex patterns.
Agentic AI could support continuous monitoring by completing several connected tasks, such as analysing transactions, reviewing customer information, identifying inconsistencies and preparing preliminary findings for investigators.
Potential applications presented at the workshop included:
- Financial-crime risk detection;
- Source-of-wealth corroboration;
- Customer onboarding checks; and
- Preparation of suspicious transaction reports.
For source-of-wealth reviews, AI could compare customer declarations with corporate records, transaction histories and other reliable information. It could then highlight gaps or inconsistencies requiring human attention.
When supporting suspicious transaction reports, the technology could organise account activity, identify connected parties and prepare a preliminary narrative for review.
Governance remains essential
The HKMA has encouraged banks to adopt proven technology at scale while exploring emerging tools. It warned that institutions risk being overwhelmed by increasingly sophisticated financial crime and excessive alerts from inefficient systems if they fail to modernise.
However, agentic AI should assist rather than replace accountable human decision-making. Banks will still need appropriate controls over:
- Data privacy and security;
- Accuracy and reliability;
- Human review and approval;
- Explainability and audit trails; and
- Model testing and ongoing monitoring.
Regulated institutions remain responsible for decisions and reports produced with the assistance of AI, even where the technology is supplied by an external vendor.
Insights from the workshop and technology demonstrations will inform future HKMA guidance. This suggests Hong Kong is preparing for wider adoption of agentic AI in AML/CFT, while recognising that its value will depend on disciplined implementation and effective human oversight.
