Hong Kong Explores AI Agents to Trace Crypto Crime as Banks Expand Adoption

10 Min Read
Hong Kong Explores AI Agents to Trace Crypto Crime as Banks Expand Adoption

Connecting crypto wallets to the people behind them

Hong Kong is exploring artificial intelligence agents that could connect blockchain transfers with bank customer records and scam intelligence, as regulators confront a crypto crime market in which illicit addresses received at least $154 billion in 2025.

Contents
  1. Connecting crypto wallets to the people behind them
  2. What an AI investigator could do
  3. Why stablecoins are central to the effort
  4. A cautious approach as licensed stablecoins develop
  5. AI adoption is broad, but autonomous tools are less common
  6. Testing agents before giving them authority
  7. Project SAGE asks who authorised the agent
  8. Intelligence sharing connects banks, platforms and police
  9. How the regulatory push has developed
  10. Key Points

Raymond Chan, the Hong Kong Monetary Authority's executive director for enforcement and anti-money laundering, outlined the idea at a regulatory technology workshop on November 17, 2026. The proposed tools could help banks spot suspicious movements between cryptocurrencies and conventional money, identify the people controlling wallets and support faster action before funds disappear.

The proposal arrives alongside evidence of growing interest in AI across Hong Kong finance. A Hong Kong Institute for Monetary and Financial Research study released the same day found that 90% of surveyed financial institutions had adopted or planned to adopt generative AI, up from 75% in 2025. However, only 13% had implemented AI agents, which can carry out a sequence of tasks with limited human supervision.

Chan's central problem is one of speed. A bank may see money moving across a public blockchain, a shared digital transaction record, yet still struggle to identify the person behind the address receiving it.

Chan explained the challenge facing investigators:

"The core challenge obviously lies in linking wallet activities to real-world identities quickly enough to prevent fund dissipation," Chan said.

The HKMA has announced neither a deployment date nor a requirement for banks to use the proposed agent. Further development and testing are the next steps, while the authority also examines how to verify an agent's identity, permissions and responsibility for its actions.

Advertisement

What an AI investigator could do

Chan envisaged combining blockchain analytics, customer identification records, intelligence about scams and monitoring of suspicious wallets within an AI agent. Tools supporting rapid fund freezes could also form part of that arrangement.

A conventional monitoring system often generates an alert for a person to investigate. An AI agent could take several preparatory steps itself: gather transaction records, compare customer details, check sanctions lists and assemble a case for an investigator. Generative AI mainly produces content, such as summaries or answers; an agent can also plan and perform tasks using connected systems.

Rapid switches between crypto and conventional currencies are one pattern Chan identified. Criminals may move money through several transactions to make its origin harder to follow, a process known as layering. Looking at bank records and blockchain transfers together could help an investigator follow a trail that crosses both systems.

Hong Kong is also researching and piloting decentralized identity tools to help establish who controls a wallet. Blockchain addresses generally appear as strings of characters rather than names. Linking an address to verified identity information could give banks a clearer basis for investigating suspicious activity.

Why stablecoins are central to the effort

Chainalysis's 2026 Crypto Crime Report, cited by Chan, estimated that illicit crypto addresses received at least $154 billion in 2025, an increase of 162% from the previous year. Activity involving sanctioned entities drove much of that increase. Stablecoins accounted for 84% of the total, while illicit activity represented less than 1% of the crypto transaction volume measured in the analysis.

Stablecoins are digital tokens designed to maintain a steady value, usually against a conventional currency. Their relatively stable price and direct transfers between wallets make them useful for payments. Those same features also attract criminals seeking to move funds quickly.

Digital assets can travel across borders and platforms rapidly, and transfers can be difficult to reverse. Investigators therefore need both a traceable transaction history and timely cooperation from organisations able to identify customers or restrict access to funds.

Advertisement

A cautious approach as licensed stablecoins develop

The HKMA granted its first stablecoin issuer licences to HSBC and Anchorpoint Financial on April 10, selecting them from 36 applicants. At the time, HKMA Chief Executive Eddie Yue expected regulated Hong Kong stablecoins to launch in the middle to second half of 2026.

The HKMA's stablecoin licensing announcement, dated April 10, 2026, named the licensees as Anchorpoint Financial Limited and The Hongkong and Shanghai Banking Corporation Limited. The licences were granted under the Stablecoins Ordinance and took effect that day. According to their business plans, the licensees intended to complete preparations and launch business in the coming few months.

Yue described the granting of licences as "an important milestone for the development of digital assets in Hong Kong" and said the regulatory regime would allow issuers to apply innovative technologies "while ensuring robust user protection and effective risk management".

Chan said the authority was taking a cautious initial approach to stablecoins, putting investor protection and controls against financial crime first. He also described the regulator as beginning to put its surveillance framework into operation.

The licensing programme and the proposed monitoring tools address connected needs. Hong Kong wants regulated digital money to become usable, while banks and issuers need ways to recognise suspicious activity and respond promptly. Identity checks, transaction analysis and cooperation with law enforcement each cover a different part of that work.

AI adoption is broad, but autonomous tools are less common

The HKIMR's AI adoption report, titled From Assistance to Autonomy: A.I. Adoption, Governance, and Compliance in Hong Kong's Financial Services Industry, draws on survey work and interviews conducted between January and July 2026. The survey itself ran from January to March and included 39 financial institutions and eight technology providers.

Current or planned generative AI adoption rose by 15 percentage points, from 75% in 2025 to 90% in 2026. For agentic AI, 13% of financial institutions had implemented the technology and 56% planned to do so. The figures show substantial interest, with most prospective users still at the planning stage.

Within regulatory compliance, 86% had adopted or planned to adopt generative AI. More than 70% expressed interest in exploring agents over the next three to five years. Banks nevertheless raised concerns about security and explainability, meaning whether staff can understand and explain how an AI system reached a decision.

One bank described using generative AI to investigate possible money laundering through transaction records, images and customers' digital activity. It reported faster and more accurate investigations, but gave no detailed performance figures. The HKMA has supported banks through technical assistance and workshops.

Advertisement

Testing agents before giving them authority

Abdul Rafay Gadit, co-founder of ZIG Finance, said banks should test agents against past fraud cases and run them alongside existing monitoring systems before allowing them to operate independently.

Gadit argued that banks should measure both detection quality and the time needed to investigate:

"Compare what it catches, what it misses and how long investigations take. Fewer alerts only helps if you're still catching suspicious activity," he said.

Such testing would let banks compare an agent's findings with known cases and existing controls. A system that produces fewer alerts may reduce work for investigators, but banks still need to measure whether it catches suspicious transactions and how often it flags legitimate customers.

Gadit said agents could gather records, check sanctions lists and prepare investigation files. Decisions to freeze accounts, reject payments or report suspicious activity should remain with bank staff, he said. For customers, that boundary matters because a monitoring decision can affect access to money or delay a payment.

Project SAGE asks who authorised the agent

The HKMA launched Project SAGE on November 2, 2026. Its full name is Securing Agentic Finance with Governance and Enablement. The project examines practical uses of financial agents alongside the safeguards needed when software can act on someone's behalf.

The official announcement sets out four areas: testing potential applications, verifying agent identities and delegated authority, developing standards that let systems work together, and examining legal and regulatory arrangements. Delegated authority means the permission a customer or institution gives an agent to perform specified actions.

The potential applications extend beyond crime detection to everyday payments, business procurement, treasury management and payments between machines for digital services. The HKMA also established an Agentic Finance Expert Group bringing together financial institutions, technology providers, professional services firms and academia.

Identity checks apply to people as well as software. Dominik Clemente, a research engineer at World Foundation, said banks should verify who is approving a flagged transaction, especially when approval happens remotely.

Clemente described the practical concern:

"If I'm asking a colleague to sign off on a flagged wire over Zoom, for example, I want to know it's really them on camera," he said.

Advertisement

Intelligence sharing connects banks, platforms and police

Hong Kong's Virtual Asset Intelligence Taskforce brings together the police, the Customs and Excise Department, the HKMA, the Securities and Futures Commission, licensed crypto platforms and stablecoin issuers. Chan said the group helps speed up responses to crime and improve information sharing.

A bank may hold a customer's identity records, while a crypto platform has information about a wallet and police possess intelligence about a scam. Bringing those pieces together can help investigators act before assets move again. Chan also pointed to the Financial Action Task Force's focus on digital assets as it works to align financial crime rules across jurisdictions.

Industry cooperation is developing too. TRON, Tether and Chainalysis have previously formed an alliance to fight crypto crime using AI. Hong Kong's approach combines technology trials with cooperation among institutions that hold different parts of a financial trail.

The risks extend beyond crime monitoring. Federal Reserve Governor Michael Barr warned on November 16 that AI agents could accelerate bank runs by moving deposits without human approval. He said banks had already encountered agents attempting to act on accounts without being able to verify customer consent. That warning makes permission checks a practical concern for any bank considering greater autonomy.

How the regulatory push has developed

The HKMA's financial crime circulars show that its AI work predates the latest crypto proposal. The sequence has moved from transaction monitoring guidance towards wider adoption and governance for agents.

  • April and September 2024: The HKMA listed a thematic review of transaction monitoring and AI, followed by guidance on using AI to monitor suspicious activities.
  • November 19, 2025: The authority issued a circular supporting AI adoption against money laundering and terrorist financing.
  • June 22, 2026: A further circular addressed AI adoption in fighting financial crime.
  • October 15, 2026: The HKMA announced its Accountability, Resilience and Trust framework for responsible AI.
  • November 2, 2026: Project SAGE launched to examine agentic finance applications and safeguards.
  • November 17, 2026: Chan outlined the crypto monitoring concept, and the HKIMR released its adoption study.

The October responsible AI framework followed sandbox trials in which more than 60% of use cases explored agents able to plan and execute tasks within authorised limits. Those trials covered applications including fraud detection and financial risk monitoring.

November's wider technology agenda also included post-quantum cryptography, tokenization and payments across borders. The crypto crime proposal gives the agentic AI work a concrete challenge: helping banks connect transaction trails with identities quickly enough for investigators to intervene, while keeping authority and accountability explicit.

Key Points

  • The HKMA is exploring AI agents that combine blockchain analysis, bank customer records and scam intelligence.
  • Chainalysis estimated at least $154 billion reached illicit crypto addresses in 2025, with stablecoins accounting for 84%.
  • Hong Kong's financial sector survey found 13% had implemented AI agents and 56% planned to do so.
  • Project SAGE examines agent identity, permissions, technical standards and legal responsibility.
  • No deployment deadline or mandatory adoption requirement has been announced for the proposed crypto monitoring agent.
Share This Article

You May also Like