Hang Seng Bank Joins Hong Kong GenA.I. Sandbox with Two Use Cases
Hang Seng Bank announced that two of its use cases have been selected for Hong Kong's GenA. I. Sandbox++ programme, jointly launched by the HKMA, SFC, IA and MPFA, with support from Cyberport. The bank will test an AI reviewer for retail banking and generative AI-assisted account opening and due diligence for commercial banking. In earlier synthetic-data trials, extracted data matched reference data with over 90% accuracy.
Hang Seng Bank announced that two of its use cases have been selected for participation in the Generative Artificial Intelligence (GenA. I.) Sandbox++ programme, jointly launched by the Hong Kong Monetary Authority, the Securities and Futures Commission, the Insurance Authority and the Mandatory Provident Fund Schemes Authority, with support from Hong Kong Cyberport Management Company Limited. The two use cases aim to enhance customer experience and operational efficiency while strengthening safeguards against AI-related risks. A key focus of the programme is testing "AI versus AI," whereby AI systems review, challenge and verify the output of another AI system, supporting better governance, auditing and human oversight.
In retail banking, Hang Seng will test an "AI reviewer" solution that reviews content generated by generative AI chatbots before responding to customer enquiries, thereby strengthening safeguards for customer-facing chatbots and delivering more accurate, appropriately toned and compliance-aligned responses. In commercial banking, Hang Seng will pilot the use of generative AI to support customer account opening and due diligence, examining the authenticity and legality of corporate information across multiple dimensions. This use case aims to verify corporate information, detect risk alerts at an early stage, and generate auditable investigation summaries to accelerate the account opening process, while ensuring that final decisions are made by human oversight.
Both pilot use cases are conducted under stringent customer data protection measures, including the principle of data minimisation, the use of public or synthetic data, and monitoring within a controlled environment. This participation builds on the outcomes of the first phase of the GenA. Sandbox, which previously conducted a proof of concept for fraud investigation automation using synthetic data that contained no real personally identifiable information. In the evaluation results of the synthetic test cases, the extracted data matched the reference data with over 90% accuracy. Hang Seng continues to share practical experience in responsible testing, prioritising customer interests and prudent regulation.