Hong Kong’s financial regulators have selected 36 use cases for the first cohort of an expanded generative artificial intelligence sandbox, with the projects set to examine how agentic AI can be used across regulated financial services.
The Hong Kong Monetary Authority said on August 27 that the cohort involves 30 financial institutions and 27 technology partners. The projects were chosen from nearly 100 proposals based on factors including innovation, technical complexity and potential industry value.
Cohort focuses on agentic AI
The GenA.I. Sandbox++ will concentrate on systems that can perform or coordinate tasks with a degree of autonomy, rather than limiting their use to generating content. Regulators said the tests will cover end-to-end processes such as customer onboarding, payments, insurance claims and customer interactions.
The pilots will also explore how one AI system might oversee actions taken by another. This extends the earlier sandbox’s “AI versus AI” approach, which considered the use of AI to review or manage risks created by AI applications.
The selected institutions will be onboarded to a platform managed through Cyberport’s Artificial Intelligence Supercomputing Centre. Technical trials are expected to begin later in 2026, according to the HKMA. Participation in the sandbox does not by itself mean that the tested systems have regulatory approval for commercial deployment.
HKT and Ant Bank outline pilot plans
Two participants separately disclosed projects that illustrate the cohort’s focus on agent-led transactions and financial risk management.
HKT Payment, the financial-services arm of Hong Kong telecommunications group HKT, said it will work with Red Date Technology on an “Agentic ID” framework for registering and verifying AI agents that initiate payment activity. The proposed framework would use decentralized identifiers and verifiable credentials to link an agent to a verified person or business.
According to HKT’s announcement, the pilot will examine payment flows including wallet top-ups, peer-to-peer transfers and transactions between institutions. The company said the framework is intended to improve accountability and auditability while using zero-knowledge proofs to verify information without exposing the underlying data. These capabilities remain objectives for the trial rather than demonstrated results.
Ant International’s Hong Kong-based digital bank Ant Bank and embedded-finance provider Bettr have meanwhile been selected for a liquidity-risk management pilot. The companies plan to use Ant International’s Falcon Time-Series Transformer 2.0 forecasting model to support daily cash-flow forecasts and treasury planning.
Bettr will support the model’s deployment for Ant Bank’s treasury team. The participants said the project is intended to develop practices for AI-assisted liquidity planning in financial services that operate around the clock. Forecast accuracy and operational benefits claimed by the companies will need to be established during testing.
Regulators widen the testing ground
The HKMA, Securities and Futures Commission, Insurance Authority and Mandatory Provident Fund Schemes Authority launched Sandbox++ in March with Cyberport. The expanded program covers banking, securities and capital markets, asset and wealth management, insurance, mandatory provident funds and stored-value facilities.
Participating institutions receive technical support, access to computing resources and guidance from their respective regulators while testing projects in a controlled setting. The program’s stated areas of interest include risk management, fraud prevention and customer experience.
Hong Kong began the narrower GenA.I. Sandbox for banks in 2024. The expanded cohort brings more parts of the financial sector into a common testing environment at a time when institutions are considering AI systems that can initiate actions, not only assist human operators.
Regulators said insights from the trials will inform their engagement with the industry. They have not provided a timetable for approving any specific system for wider use, and each proposed deployment would remain subject to the applicable licensing, governance, privacy and risk-management requirements.
Featured image: Daniam Chou on Unsplash