What to Know
- Hong Kong’s central bank is encouraging banks to adopt artificial intelligence to detect financial crime linked to digital assets.
- The Hong Kong Monetary Authority is exploring how AI agents could connect blockchain transactions with customer records to identify suspicious activity more quickly.
- Illicit cryptocurrency addresses received at least $154 billion in 2025, up 162% from the previous year, with stablecoins accounting for 84% of that total.
- The same crypto crime data placed illicit activity at less than 1% of the crypto transaction volume measured in its analysis.
- Market specialists say AI agents should be tested against past fraud cases and operated alongside existing monitoring systems before being allowed to act independently.
- Banks are weighing the benefits of faster investigations against concerns over security, accountability and explainability.
- The HKMA launched Project SAGE on Nov. 2 to examine identity, authority and responsibility issues around financial AI agents.
Hong Kong Sharpens Its AI Focus in Crypto Crime Fight
Hong Kong’s central bank is stepping up pressure on banks to use artificial intelligence in the fight against financial crime, with digital assets and stablecoins becoming a bigger part of the monitoring challenge. The Hong Kong Monetary Authority is focusing on how banks can use AI agents to identify suspicious transfers, connect blockchain activity with customer information and respond before funds are moved beyond easy recovery.
The effort reflects a growing problem for regulated financial institutions. Blockchain networks can make transfers visible, but that does not automatically tell banks who controls a wallet, whether a transaction is connected to a sanctioned entity or whether a rapid movement of funds is part of an attempt to disguise the origin of money. For banks, the race is not only to see the transaction, but to understand it quickly enough to act while the money is still traceable.
Raymond Chan, the Hong Kong Monetary Authority’s executive director for enforcement and anti-money laundering, said the regulator is working to accelerate AI adoption among banks. That includes examining how AI agents could combine blockchain transfers with customer records to detect suspicious patterns that may otherwise be missed or reviewed too slowly through conventional compliance processes.
Stablecoins Sit at the Center of the Risk Debate
The urgency is being amplified by the scale of illicit activity tied to digital assets. Illicit cryptocurrency addresses received at least $154 billion in 2025, up 162% from the previous year. Stablecoins accounted for 84% of the total, underscoring how widely used blockchain-based dollar substitutes have become in both legitimate and suspicious payment flows.
At the same time, illicit activity was placed at less than 1% of the crypto transaction volume measured in the relevant analysis. That distinction matters for banks and regulators. The policy challenge is not simply to treat all digital asset activity as suspicious, but to build monitoring systems that can separate routine customer activity from high-risk patterns with sufficient accuracy and speed.
Stablecoins are especially important because they can move quickly across blockchain networks while maintaining a value profile that is easier to use for settlement than more volatile tokens. That utility makes them attractive for payments, trading and treasury activity, but it can also complicate anti-money laundering work when funds move through multiple wallets, platforms or jurisdictions in a short period.
The Identity Gap Behind Blockchain Monitoring
Banks can examine transactions recorded on public blockchains, but identifying the individual or entity behind a wallet remains a more difficult task. Chan described the core challenge as linking wallet activity to real-world identities quickly enough to prevent fund dissipation. In practice, that means banks need systems that can interpret transaction behavior, match it against customer records and surface risk indicators before a questionable transfer becomes much harder to recover.
AI agents could help by reviewing transaction records, identifying connections across accounts, scanning sanctions data and organizing evidence for compliance teams. They may also be able to flag patterns such as rapid switches between cryptocurrencies and conventional money, which can be used to obscure the source of funds. Identity tools designed to help establish who controls a wallet are also being piloted.
For financial institutions, this is a major operational shift. Traditional monitoring systems often produce large volumes of alerts, many of which require human review. AI could reduce noise, highlight more relevant cases and provide investigators with richer context. But fewer alerts are only useful if the system continues to catch genuinely suspicious activity rather than simply filtering out inconvenient complexity.
Experts Urge Parallel Testing Before AI Agents Act Alone
Abdul Rafay Gadit, co-founder of ZIG Finance, said banks should test AI agents against past fraud cases and run them alongside existing monitoring systems before allowing them to operate independently. In his view, banks need to compare what AI catches, what it misses and how long investigations take. The key point is that reducing the number of alerts is not a success if important risks are being overlooked.
That cautious approach reflects a broader concern across the banking sector. AI agents are different from passive analytics tools because they can perform several tasks with limited human supervision. In a financial crime setting, that may include gathering documents, enriching transaction histories, checking sanctions lists and preparing cases for investigators. The more capable these tools become, the more important it is to define where automated assistance ends and human judgment begins.
Gadit said decisions to freeze accounts, reject payments or report suspicious activity should remain with bank staff. That line is likely to be central to how banks adopt AI in regulated environments. Automation can accelerate the investigative process, but final decisions that affect customer access to funds, payment execution or regulatory reporting carry legal, reputational and operational consequences.
HKMA Support Meets Governance Concerns
The Hong Kong Institute for Monetary and Financial Research released a report Tuesday describing HKMA technical support and workshops designed to help banks deploy AI. The report included an example of one bank using generative AI to investigate possible money laundering by reviewing transaction records, images and customers’ digital activity. The bank reported faster and more accurate fraud investigations, though details were not provided.
The survey behind the report found that 90% of surveyed financial institutions had adopted or planned to adopt generative AI. Across business use cases, 13% had implemented AI agents, while 56% planned to do so. The survey was conducted from January to March and included 39 financial institutions and eight technology providers.
Those findings point to rapid adoption, but not without hesitation. Banks raised concerns over security and the ability to explain AI-made decisions. Explainability is especially important in compliance because institutions may need to show regulators, auditors and customers why a payment was stopped, why an account was flagged or why a case was escalated.
Security is another concern. AI systems that handle customer records, transaction data and compliance workflows can become sensitive operational infrastructure. If poorly governed, they may create new risks even as they reduce old ones. Banks will need controls around data access, model behavior, audit trails, escalation procedures and human approvals.
Remote Approvals Add Another Layer of Risk
Dominik Clemente, a research engineer at World Foundation, said banks should verify that the person approving a flagged transaction is who they claim to be, especially when approvals happen remotely. He pointed to the example of asking a colleague to sign off on a flagged wire over Zoom and needing confidence that the person on camera is genuine.
That concern fits into a wider discussion about identity in the age of AI. As fraudsters use more advanced tools, banks may need stronger methods to verify not only customers, but also internal approvals and agent authority. The question is no longer only whether a transaction looks suspicious, but whether the people and systems authorizing action are legitimate.
Federal Reserve Governor Michael Barr warned Monday 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. The warning highlights why financial AI is not only an anti-crime tool, but also a potential source of systemic and operational risk if deployed without proper safeguards.
Project SAGE Targets Identity and Accountability
The HKMA separately launched Project SAGE on Nov. 2 to examine how financial AI agents should prove their identity and authority to act, as well as who should be responsible when things go wrong. The project places accountability at the center of Hong Kong’s AI finance agenda, recognizing that banks must be able to distinguish between authorized automation and unauthorized action.
For crypto-linked financial crime, the stakes are high because funds can move quickly across wallets and platforms. If AI helps banks identify suspicious activity sooner, it could improve the odds of intervention before funds disperse. But if AI systems act too aggressively or without clear oversight, they could disrupt legitimate customers, create legal disputes or expose banks to new forms of fraud.
Hong Kong’s approach suggests that AI adoption in banking will be shaped by both innovation and restraint. Regulators want faster detection, stronger identity tools and better links between blockchain data and customer records. Banks, meanwhile, need systems that are accurate, explainable and governed by clear human decision-making. The outcome could influence how other financial centers think about digital asset monitoring as stablecoins and blockchain-based payments become more embedded in global finance.
Frequently Asked Questions (FAQs)
Why is Hong Kong pushing banks to use AI against crypto crime?
Hong Kong’s central bank wants banks to detect suspicious digital asset transfers more quickly as stablecoins and other crypto assets make money movement faster and harder to trace through traditional systems.
What role could AI agents play in bank compliance?
AI agents could review blockchain transactions, compare activity with customer records, check sanctions lists, identify suspicious patterns and prepare cases for human investigators.
Why are stablecoins a focus for regulators?
Stablecoins accounted for 84% of the illicit crypto value cited for 2025, making them a major focus for banks trying to monitor suspicious blockchain-based transfers.
How much did illicit crypto addresses receive in 2025?
Illicit cryptocurrency addresses received at least $154 billion in 2025, up 162% from the previous year.
Does this mean most crypto activity is illicit?
No. The relevant analysis placed illicit activity at less than 1% of the crypto transaction volume measured, meaning the challenge is identifying risky activity without disrupting legitimate users.
What are the main risks of using AI in banking compliance?
The main risks include security weaknesses, poor explainability, incorrect decisions, unclear accountability and excessive reliance on automated systems without human oversight.
Should AI be allowed to freeze accounts automatically?
Some market specialists argue that AI can support investigations, but decisions to freeze accounts, reject payments or report suspicious activity should remain with bank employees.
What is Project SAGE?
Project SAGE is an HKMA initiative launched on Nov. 2 to examine how financial AI agents should prove identity and authority, and who should be accountable when problems occur.
