Hong Kong Studies AI for Crypto Crime Monitoring

Hong Kong’s monetary authority is exploring whether agentic artificial intelligence can help financial institutions identify crypto-related financial crime by examining blockchain activity alongside customer transactions. The initiative remains exploratory, with no new rule announced, but it places AI monitoring alongside the city’s broader work on digital-asset oversight and information sharing.
What happened
The Hong Kong Monetary Authority, or HKMA, is exploring the development of agentic AI that could analyze both on-chain transfers and off-chain customer transactions, according to an HKMA speech by Raymond Chan, Executive Director for Enforcement and AML. The intended role is to flag anomalies, including rapid fund layering through switches between cryptocurrencies and fiat currencies.
Agentic AI generally refers to software that can carry out multistep tasks. TokenPost reported that Chan outlined a possible use case in which an AI agent could help banks detect suspicious crypto activity by connecting blockchain transfers with customers’ off-chain transactions. The concept would bring together blockchain analytics, customer due-diligence information and transaction monitoring rather than treating those data points as separate compliance exercises.
Bitcoin.com News reported that Chan said agentic AI could become relevant to the difficulty of applying traditional anti-money-laundering and counter-terrorist-financing measures to digital assets. He pointed to the transaction trails left when digital assets move as a potential advantage for monitoring, despite the complexity of following value across different assets and payment channels.
The HKMA’s stated focus is not simply on identifying an individual transfer. It is on detecting patterns that may signal layering, where funds move quickly through multiple conversions and channels. Such patterns can be difficult to assess when a bank sees only its own customer activity and does not have a complete view of related on-chain movements.
As of Thursday, the proposal is a study of possible monitoring tools, not a requirement for banks or digital-asset firms to deploy AI agents. The distinction matters because the technical, governance and legal arrangements around such systems have not been set out as a new regulatory framework.
Why it matters
The initiative reflects a practical compliance question for institutions that serve customers using digital assets. On-chain data may show where and when transfers occur, while bank records and customer due-diligence files can offer a separate view of account behavior. Linking those views could help surface activity that appears ordinary in either dataset alone.
It also raises questions about how far automated analysis should influence financial-crime decisions. An anomaly flag is not, by itself, a conclusion about unlawful conduct. Banks and authorities would need clear processes for reviewing alerts, determining whether escalation is appropriate and assigning responsibility for actions taken on the basis of an AI-generated assessment.
CoinDesk reported that a Hong Kong Institute for Monetary and Financial Research survey of 39 financial institutions and 8 technology providers found that 90% had adopted or planned to adopt generative AI. The survey, conducted from January to March, found that 13% had implemented AI agents and 56% were planning to do so. Those figures indicate that financial institutions are already considering AI agents beyond the narrower crypto-monitoring use case described by Chan.
Chan cited the 2026 Chainalysis Crypto Crime Report in the HKMA speech, saying illicit addresses received over $150 billion in 2025 and that stablecoins accounted for 84% of that volume. The amount cited elsewhere differs slightly: CoinDesk reported a Chainalysis figure of at least $154 billion and said it was up 162% from the prior year. The reporting agrees on the broader point that the AI work is exploratory and that no new rule has been announced.
Background and context
Banks in Hong Kong are expanding into tokenised bonds, funds, tokenised deposits and stablecoins. That expansion increases their exposure to financial-crime risks associated with digital assets, while also increasing the need to integrate digital-asset activity into existing compliance systems.
Digital assets can combine speed, global reach, pseudonymity and limited reversibility. Funds may move internationally and bypass intermediaries, while recovery can become extremely difficult once assets have been transferred through multiple routes. These characteristics are why the Financial Action Task Force has made digital assets a key priority.
Hong Kong has also established the Virtual Asset Intelligence Taskforce, or VAIT. The Block reported that VAIT connects the Police, the Customs and Excise Department, the HKMA, the Securities and Futures Commission, licensed virtual asset service providers and stablecoin issuers. Forklog similarly reported that the taskforce is intended to share information and coordinate detection and prevention of digital-asset crimes.
Identity is another part of the policy discussion. TokenPost reported that the HKMA is exploring and piloting decentralized identity tools that could link a wallet to a verified person. Chan also described the authority’s initial approach to stablecoins as cautious, with investor protection and financial-crime controls among the priorities.
CoinDesk reported that the HKMA launched Project SAGE to examine how financial AI agents should establish identity and authority, as well as who should be accountable when problems occur. That work addresses a separate but connected issue: monitoring systems may become more capable only if the institutions using them can verify what an AI agent is allowed to do and who bears responsibility for its output.
What to watch
Key questions include what data an agentic AI system would be permitted to use, how on-chain and off-chain information would be linked, and what safeguards would apply to customer information. The HKMA has described a potential analytical capability, but has not detailed a required technical architecture or implementation timetable.
It will also be important to see how any AI monitoring effort interacts with VAIT’s information-sharing role, decentralized identity pilots and the surveillance framework mentioned by Chan. The division of responsibility between banks, licensed platforms, stablecoin issuers and public authorities remains central to any system designed to respond to suspicious activity.
Finally, Project SAGE’s focus on identity, authority and accountability leaves open questions about governance. Financial institutions may need clarity on how agent outputs are reviewed, when human intervention is required and how responsibility is determined if an automated process produces an incorrect or incomplete alert.
