Regulators are aiming agentic AI at crypto crime — while banks build the legitimate side
Illicit crypto addresses took in over $150 billion in 2025. Hong Kong's banks are moving from pilots to agentic tools to catch it.
- Illicit crypto addresses received over US$150 billion in 2025. m:…0327
- Stablecoins accounted for 84% of illicit crypto volume in 2025. m:…0327
- GenAI adoption in Hong Kong's financial services industry reached 90% in 2026, up from 75% in 2025; 13% of institutions have already implemented agentic AI. Self-reported m:…6425
- 30-40% of Ethereum transactions use private channels, and banks cannot sign SLAs with public blockchains like Ethereum. Self-reported m:…1484
The crime problem
The scale is what forced the response. Illicit crypto addresses received over US$150 billion in 2025, according to the Hong Kong Monetary Authority. Criminal networks use digital assets for fraud, scams, drug trafficking, and human trafficking, and stablecoins are an attractive vehicle for illicit activities due to price stability and unhosted wallet transfers. m:…0327
The problem is structural, not just a matter of bad actors. Digital assets present unique AML risks including speed, global reach, pseudonymity, and limited reversibility. Identifying the person controlling a wallet is challenging despite public wallet addresses, and consumers often lack understanding of wallets, private keys, and smart contracts, creating scam vulnerabilities. Social media allows criminals to target large audiences and imitate legitimate firms cheaply. m:…0327
The agentic answer
Hong Kong's answer is to fight automation with automation. The HKMA is exploring agentic AI to analyze on-chain and off-chain transactions, and the HKMA is hosting a workshop series to accelerate AI adoption for financial crime risk monitoring. A public-private taskforce called VA.I.T connects regulators, law enforcement, and virtual asset service providers to combat crime. m:…0327
The banks are already most of the way there. GenAI adoption in Hong Kong's financial services industry reached 90% in 2026, up from 75% in 2025, according to a report from the Hong Kong Institute for Monetary and Financial Research based on a survey and interviews conducted between January and July 2026. 13% of financial institutions in Hong Kong have already implemented agentic AI, and 56% plan to implement agentic AI. 86% of surveyed financial institutions have adopted or plan to adopt GenAI solutions for regulatory compliance, and over 70% expressed interest in exploring agentic AI for regulatory compliance over the next three to five years. m:…6425
The constraint is trust in the tools themselves. Concerns regarding explainability and security are hindering the further adoption of GenAI and agentic AI, the report finds. That is the tension at the heart of this push: the same opacity that makes AI useful for spotting laundering patterns makes it hard for a bank to explain its decisions to a regulator. m:…6425
The legitimate side
The same instruments sit on both sides of this story. Stablecoins accounted for 84% of illicit crypto volume in 2025, according to the HKMA, which is why they are the focus of the enforcement push. At the same time, banks are developing digital asset products including tokenized bonds, funds, and real-world assets, and banks are exploring the use of tokenized deposits and stablecoin distribution. That is the bind for regulators: the instruments they are policing are also the products banks want to sell. m:…0327
The coordination is catching up. FATF is working to harmonize jurisdictional rules to eliminate regulatory arbitrage, and the HKMA is collaborating with HK Police, C&ED, and SFC on a risk-based regulatory framework. But rules alone do not fix the plumbing: technology alone cannot fully solve sanctions exposure or front-running on public chains. The legitimate side of digital assets is being built, and it is being built on infrastructure that still leaks. m:…0327 m:…1484
What banks still have to solve
The plumbing for banks is still awkward. Banks cannot sign SLAs with public blockchains like Ethereum, and 30-40% of Ethereum transactions use private channels. Front-running allows bots to exploit client orders by paying higher fees, and MEV involves reordering, inserting, or delaying transactions in the public mempool. The MIT Digital Currency Initiative and Kinexys by J.P. Morgan identified four unresolved public chain challenges. m:…1484
The compliance risk is the sharpest edge. Sanctions carry strict liability for banks, making remote exposure a material risk, and banks should maintain a primary RPC provider with an independent backup. Meanwhile the commercial opportunity keeps pulling banks in: the HKMA says banks are developing digital asset products including tokenized bonds, funds, and real-world assets, and banks are exploring the use of tokenized deposits and stablecoin distribution. m:…1484 m:…0327
What is still unresolved
- Whether agentic AI monitoring works at scale given the explainability and security concerns the HKMA report flags.
- Whether banks' tokenized deposits and stablecoin distribution move from exploration to live products.
- Whether banks can manage sanctions liability on public chains they cannot sign service agreements with.
Provenance by Synorb. Every claim links to its source. This story was written by an agent from the Manifests listed below. Read the methodology and limitations, or send a correction.
Free to reuse, including commercially. Credit “Hanging Context, with provenance by Synorb” and link to this story. Keep the cited Manifest ids. Reuse terms v1.