Hong Kong financial regulators enhance their practical artificial intelligence with select 36 use cases for the first cohort of the GenAI Sandbox++. Projects to test span across agent-initiated commerce, payments, customer onboarding and liquidity management.
The initiative brings together the Hong Kong Monetary Authority (HKMA), Securities and Futures Commission (SFC), Insurance Authority and Mandatory Provident Fund Schemes Authority, working with Cyberport. The institutions in question selected projects from 30 financial institutions and 27 technology partners. Technical trials of those prototype initiatives are expected to begin later this year. The projects were assessed based on their level of innovation, technical complexity and potential value to the financial industry.
Some of the chosen pilots will test end-to-end financial processes, while others will examine whether AI can oversee actions taken by other AI applications. The participants will use a platform managed by Cyberport’s AI Supercomputing Centre, giving firms a controlled environment for testing more autonomous systems with regulatory involvement from an early stage.
“We look forward to leveraging the profound insights from these trials to empower the broader market transformation,” said Julia Leung, Chief Executive Officer of the SFC
Eddie Yue, Chief Executive of the HKMA, noted: “We are delighted to foster more innovative partnerships between a wide range of financial institutions, in addition to banks, and technology companies. These partnerships are unlocking the potential of Agentic A.I. to create new opportunities.”
Visa to Test AI Agents Making Commercial Payments
One of the most prominent pilots brings Visa and Bank of China (Hong Kong) together to test agent-initiated commerce.
The proof of concept will connect BOCHK’s AI agent within the BOC Pay+ ecosystem to services including OpenRice and China Mobile Hong Kong. The objective is to allow an AI agent to discover services, initiate purchases and complete transactions with consumer consent inside a risk-controlled environment.
Visa will provide the payment infrastructure, controls and connectivity required for those transactions, with its Visa Intelligent Commerce technology that has tokenized Visa credentials to support agent’s authorization serving as the framework for the pilot. All transactions will happen in a risk-controlled sandbox environment to avoid any incidents with payment infrastructure involved. Visa has earlier tested its agentic payment technology with several financial institutions, including a live agentic-commerce proof of concept with Nuvei, merchant technology provider Arvato Systems, and fashion brand Kings and Priests.
“Visa’s role is to provide the trusted payment infrastructure that enables AI agents to transact safely on behalf of consumers, with appropriate safeguards, transparency and user control built into every transaction, while enabling merchants to engage confidently in AI-powered commerce.”
Paulina Leong, General Manager of Visa Hong Kong and Macau
The payment card network is simultaneously developing AI for the defensive side of agentic commerce. Its updated Visa Vulnerability Agentic Harness (VVAH) cyber defense system that has undergone modifications post participation in Anthropic’s Project Glasswing is designed not only to identify vulnerabilities but also to support remediation and validation. Visa says some attack paths can now be resolved in hours rather than weeks, while fixes that fail validation can be refined without restarting the process.
“AI is compressing the time between vulnerability discovery and exploitation, which means defenders need a faster, more reliable path to action.”
Rajat Taneja, President of Technology at Visa
That combination of cybersecurity functions is becoming increasingly important as financial institutions give AI systems greater ability to interact with sensitive infrastructure.
Ant Bank Brings AI Into Treasury Management
Another sandbox project focuses on a less visible but not less potentially significant application of AI in banking: bank liquidity management.
Ant Bank and Bettr have been selected for the liquidity risk management test project. Ant Bank plans to use Ant International’s Falcon Time-Series Transformer (TST) AI Model 2.0 to improve foreign exchange and liquidity management, while Bettr’s platform technology will support its deployment for the bank’s treasury team. The project aims to generate more granular daily forecasts for real-time liquidity planning.
Ant International has reported that Falcon TST 2.0 achieved more than 93% forecast accuracy on a leading public time-series AI benchmark, although the sandbox project will provide a more relevant test of how such technology performs in an operational banking environment.
Chinese tech and e-commerce giant has been actively developing its AI ecosystem, embedding artificial intelligence technology deeply in its commercial stack, healthcare offering, and now experimenting with banking infrastructure as well.
Earlier this month, Alipay launched a product marketed as China’s first full-stack agentic commerce platform for merchants. The platform helps businesses turn existing products and services into AI-ready “Skills” and tools that agents can use to browse catalogues, book services and execute tasks. It also connects into Alipay’s Ah Bao ecosystem, which can perform thousands of everyday services and incorporates payment, identity verification, risk management and fulfilment capabilities.
Hong Kong’s sandbox arrives at a stage where banks, payment networks and merchants are beginning to explore what happens when AI can forecast liquidity, select services and actually transact. The regulatory challenge here is tough. Industry players and legislators should find a common way of making that autonomy useful without allowing it to become uncontrolled. By putting these applications into a supervised sandbox, Hong Kong is testing not just what AI agents can do, but how financial institutions can safely let them do it.
