Hong KongFinance and insurance2024–2025Hong Kong Monetary Authority (HKMA) and Cyberport
GenA.I. Sandbox for banks, first cohort
Hong Kong's GenAI sandbox found banks could cut suspicious transaction report preparation time by 30% to 80%
Problem
Banks wanted to use generative AI in risk management, anti-fraud work and customer service but lacked a safe place to test it with supervisors watching.
Approach
The HKMA and Cyberport opened a GenA.I. Sandbox in December 2024. Banks ran trials with technology partners on Cyberport's AI supercomputing centre. The first cohort ended with a symposium on 31 October 2025, a report on findings and a chatbot built only on sandbox material to limit hallucinations. A second cohort of 27 use cases from 20 banks started then.
Results
Takeaway
Banks shortened suspicious transaction and risk assessment report preparation in supervised trials. The reported savings apply to the test use cases.
Categories
- Size
- Large enterprise
- Industry
- Finance and insurance
- Business function
- Finance, risk and legalCustomer service
- Innovation type
- Process innovation
- AI technology
- Generative AI
- Adoption
- Government support program
- Evidence
- Regulator, audit or court
Sources
Source summaries
HKMA and Cyberport held a symposium with over 500 attendees to close the first sandbox cohort and launch the second. HKMA reported potential uses in risk management, anti-fraud and customer service, and introduced a chatbot that answers only from sandbox material.
The Standard reported the first-cohort findings. Suspicious transaction report preparation fell 30% to 80%, document processing dropped from a day to about five minutes, and risk assessment reports took 60% less time. 86% of users rated the system favourably, and over 70% of generated credit assessments were judged useful for reference.
Full details are in the original sources.
