Hong Kong Explores Agentic AI to Track Crypto Crime Across Banks and Blockchains

Key Takeaways
- The HKMA says agentic AI is being explored as a potential way for banks to analyze on-chain transfers alongside off-chain customer transactions.
- The proposal is not a deployed monitoring system or a new rule; the authority has framed it as one possible tool within layered safeguards.
- Hong Kong’s Virtual Asset Intelligence Taskforce already provides a public-private channel for sharing intelligence about digital-asset crime.
Hong Kong’s central bank wants banks to consider a new kind of financial-crime monitor: an AI agent able to examine blockchain transfers alongside a customer’s conventional banking activity. The Hong Kong Monetary Authority (HKMA) says it is exploring the approach, not deploying a finished system. The distinction matters. It signals where regulators see potential, while leaving questions about access, accountability and intervention unresolved.
In a 17 November speech, Raymond Chan, the HKMA’s executive director for enforcement and anti-money laundering, described agentic AI as a possible addition to banks’ monitoring capabilities. It could flag patterns such as rapid movement between cryptoassets and fiat currencies. The proposal combines data that today may sit in separate systems; it does not mean the authority has announced an autonomous enforcement tool or a requirement for every bank to adopt one.
What the HKMA is exploring
A blockchain can make transaction histories visible without making wallet owners obvious. Analysts may follow funds across public addresses, but connecting an address to a real person often depends on other evidence: customer due-diligence records, exchange account information, scam reports and lawful cooperation with service providers or authorities.
Chan’s idea is to bring several of those signals together. An AI agent could compare on-chain transfers with off-chain customer transactions and identify anomalies—for example, a fast sequence of crypto-to-fiat movements that merits human review. In principle, this could help a bank notice a pattern sooner than a process that requires analysts to switch manually among tools and datasets. The Block’s report on the speech likewise described the proposal as exploratory: the HKMA is considering what such analysis could do, rather than announcing a live product. The distinction is important because automated systems can surface a lead, but the institution still has to verify its relevance and decide what response is lawful and proportionate.
That is a monitoring use case, not proof that an AI can reliably identify a criminal or determine that a transaction is illicit. A suspicious pattern is a lead, not a verdict. A legitimate customer can have unusual activity; a sophisticated criminal can stay below simple thresholds. The HKMA’s speech stresses that no single tool offers complete protection and calls for multiple safeguards.
Why stablecoins are part of the discussion
The authority pointed to the speed, global reach and limited reversibility of digital-asset transfers. Stablecoins add another dimension: they are designed to track fiat currencies and can move between wallets around the clock. That convenience is useful for payments, but it can also make them attractive in illicit flows.
Chan cited the 2026 Chainalysis Crypto Crime Report, which estimated that illicit addresses received more than US$150 billion in 2025 and that stablecoins made up 84% of the volume it classified as illicit. Those figures are estimates from a blockchain analytics firm, not a count of every crime or a measure of all stablecoin use. The vast majority of ordinary stablecoin activity is not thereby shown to be criminal.
A data-linking problem, not just an AI problem
The difficult step is often identity. Public ledgers can show that funds moved from one address to another; they do not inherently reveal who controlled each wallet at a particular moment. An agent that joins ledger activity to bank and customer data would still depend on the quality, timeliness and permitted use of those inputs.
False matches and stale intelligence could create real costs: a customer may face a delayed transfer or an account review because a system misread context. Conversely, missing or fragmented data can let an actual risk pass unnoticed. These are familiar challenges in transaction monitoring, but greater automation can make them faster and less visible unless banks can explain why an alert was raised and how it was handled. Governance would need to cover not just the model but the complete chain: data quality, permissions, audit trails, access controls, escalation thresholds and a route for staff to correct errors. If an agent can trigger actions rather than merely recommend them, the boundary between detecting risk and restricting a customer becomes especially consequential.
Hong Kong’s existing cooperation framework
The speech places the AI exploration alongside the Virtual Asset Intelligence Taskforce (VAIT), a public-private channel involving the police, Customs and Excise Department, HKMA, Securities and Futures Commission, licensed virtual-asset providers and stablecoin issuers. Chan said the taskforce supports intelligence sharing, disruption, prevention and detection.
That institutional layer is important because technology alone cannot freeze funds or establish legal responsibility. Any action depends on the relevant institution’s authority, its own procedures and applicable law. A system that spots a pattern may help investigators coordinate, but it does not replace those decision-makers.
What banks and customers should watch next
The HKMA has not announced a launch date, technical design, participating banks or binding implementation timetable for this crypto-monitoring idea. The next meaningful developments would be concrete pilot details: what data can be combined, who can access it, how alerts are validated, how errors are challenged and what human approval is required before accounts or transfers are restricted. It will also matter whether the system only ranks cases for investigators or can initiate a freeze request, and how institutions preserve evidence when several providers and jurisdictions are involved. Those design choices determine whether automation is an analyst’s aid or a new operational risk.
For customers, the practical lesson is not that routine crypto transfers will automatically be judged by an AI. It is that financial institutions are looking for ways to connect activity across traditional and digital channels. Clear records of the source and purpose of funds can help if a legitimate transaction prompts a compliance question; no monitoring system can guarantee that a transfer will be approved or recovered.
The promise—and the limit
A well-governed agent could help analysts connect clues that are currently separated across ledgers, customer files and reports. But the HKMA’s announcement is exploratory, and the core challenge remains broader than model capability: timely data, reliable identity links, privacy and security controls, explainable alerts, and accountable human decisions. A false positive can disrupt a legitimate payment; a missed connection can leave funds moving beyond recovery. Independent validation, careful access limits and a clear record of how each alert was produced would therefore matter as much as speed. These safeguards are not optional polish around an AI feature—they are part of whether a bank can responsibly use it at all.
For now, Hong Kong has outlined a possible direction rather than a working system. Whether agentic AI improves crypto-crime monitoring will depend less on the label attached to the software than on the evidence it can use, the safeguards around its actions and the institutions responsible for acting on its warnings.
DISCLAIMER
This article is for informational purposes only and does not constitute financial advice. Cryptocurrency investments involve substantial risk and extreme volatility - never invest money you cannot afford to lose completely. The author may hold positions in the cryptocurrencies mentioned, which could bias the presented information. Always conduct your own research and consider consulting a qualified financial advisor before making any investment decisions.









