Ping An Digital Bank has completed an anti-fraud testing project in the second cohort of the Hong Kong Monetary Authority’s Generative AI Sandbox, reporting a deepfake detection accuracy of more than 96% in the trial.
The Hong Kong digital bank said on October 15 that its Anti-Fraud Strategy Platform, developed with OneConnect Financial Technology, was tested on the identification of forged or synthetically manipulated faces in the customer verification process.
The platform combines facial recognition with real-time assessment of customer selfies to help identify suspicious imagery. The bank described the work as an exploration of AI-based defenses against AI-generated fraud rather than announcing an independently certified production accuracy rate.
Ronald Iu, chief executive of Ping An Digital Bank, said the sandbox helped the institution validate the potential of AI-based deepfake detection across different scenarios. The reported accuracy reflects the testing conditions used by the project; the company did not publish a full methodology or false-positive rates in the announcement.
The project is part of a sandbox run by HKMA in collaboration with Cyberport. The regulator has been encouraging banks to test generative and agentic AI within supervised environments, including work on governance, security and fraud prevention.
On the same date, HKMA announced its ART framework for accountable, resilient and trustworthy AI adoption in banking, alongside further experimentation through the GenA.I. Sandbox++. The bank said it plans to continue applying AI to risk control and product development, but did not give a commercial deployment schedule for this particular fraud platform.