36 agentic AI use cases selected for Hong Kong sandbox trials
Hong Kong financial regulators have selected 36 use cases for the first cohort of the GenA.I. Sandbox++. The trials involve 30 financial institutions and 27 technology firms testing autonomous artificial intelligence across banking, investment, insurance, and pensions.
Autonomous systems enter sandbox testing
The selection committee evaluated nearly 100 proposals based on innovation, technical complexity, and financial sector value before approving 36 projects.
The collaborative initiative spans the Hong Kong Monetary Authority, the Securities and Futures Commission, the Insurance Authority, and the Mandatory Provident Fund Schemes Authority.
Moving beyond standard generative tools, the trials focus on agentic systems capable of executing end-to-end workflows with operational autonomy.
The scope covers customer onboarding, payment execution, insurance claim processing, and client interactions.
Furthermore, the cohort tests dynamic supervisory mechanisms where autonomous models monitor other artificial intelligence applications.
All trials will run on Cyberport’s Artificial Intelligence Supercomputing Centre platform.
Supervisory engagement across four sectors
The initiative unites 30 institutions and 27 technology partners across banking, securities, insurance, and mandatory provident funds.
“The financial industry’s exploration of agentic A.I. marks a pivotal step forward in developing a future-ready Hong Kong market, where greater system autonomy is underpinned by sound governance and clear accountability,” Securities and Futures Commission Chief Executive Julia Leung said.
The framework allows regulators to maintain early supervisory engagement while testing risk controls.
Supervised agents over unchecked automation
Moving from text generation to autonomous execution marks a major operational shift for financial supervision.
Uniting four sectoral regulators under one testing facility provides a necessary and practical framework.
Long-term success will depend on whether automated oversight can effectively prevent errors before live deployment.
Source: First cohort of GenA.I. Sandbox++
IN: