Ping An Digital Bank Achieves 96% Deepfake Detection Accuracy in HKMA GenAI Sandbox

Ping An Digital Bank has successfully completed its "Anti-Fraud Strategy Platform" use case within the Hong Kong Monetary Authority’s Generative Artificial Intelligence Sandbox. For fintech professionals, this marks a significant milestone in the "A.I. vs A.I." defense landscape, demonstrating a 96% accuracy rate in detecting deepfake-related fraud during real-time identity verification processes.

What was announced

Ping An Digital Bank (International) Limited, also known as PingAnDB, participated in the second cohort of the HKMA’s GenA.I. Sandbox in partnership with Hong Kong Cyberport Management Company Limited. The primary focus of the project was the development and validation of an "Anti-Fraud Strategy Platform" designed to counter the rising threat of deepfake technology in the banking sector.

The platform was developed in collaboration with OneConnect Financial Technology, utilizing proprietary technology from both Ping An Group and OneConnect. By integrating advanced A.I. facial recognition into the bank’s identity verification workflow, the system analyzes user selfies in real time. This allows the bank to identify and flag suspected forged or manipulated faces, effectively creating an automated defense mechanism against synthetic media.

Testing conducted within the sandbox environment yielded a detection accuracy rate exceeding 96%. This high level of precision allows the bank's risk management teams to proactively monitor and prevent fraudulent account openings or transactions. The initiative specifically explored the potential of an "A.I. vs A.I." strategy, where generative models are used to train and refine the detection capabilities of the bank’s security infrastructure across various fraud scenarios.

"Ping An Digital Bank is honoured to participate in the HKMA’s second GenA.I. Sandbox cohort. The sandbox serves as an invaluable testing ground for banks to explore financial technology. Throughout the process, we received valuable guidance from the HKMA and shared practical experience with industry peers, advancing the development of anti-fraud efforts across the sector. This project successfully validated GenAI’s potential in deepfake detection, providing deeper insights into how AI identifies forged or synthetic faces across various scenarios. Sandbox testing results demonstrated a detection accuracy of over 96%. We believe the experience gained will help strengthen the industry’s ability to counter deepfake fraud, and build a more robust defence mechanism."

Mr. Ronald Iu, Chief Executive of Ping An Digital Bank

The companies involved

Ping An Digital Bank is a digital-native financial institution operating under the umbrella of the Ping An Group, one of the world's largest integrated financial services groups. The bank focuses on leveraging the group's extensive technological ecosystem to provide digital banking services. It works closely with OneConnect Financial Technology, a leading technology-as-a-service provider for financial institutions, to implement advanced risk control and operational tools.

The project was supported by Hong Kong Cyberport Management Company Limited, a key player in Hong Kong's digital tech ecosystem. Cyberport is a wholly-owned company of the Hong Kong SAR Government and serves as a hub for over 2,000 community members, including numerous fintech startups and established technology firms. Its role in the GenA.I. Sandbox highlights its position as a facilitator for regulatory innovation and technological adoption within the region's financial sector. Together, these entities represent a concentrated effort to maintain Hong Kong's status as a global fintech hub by addressing emerging security challenges through collaborative testing and development.

What FF News has reported before

FF News has closely followed the development of the HKMA’s GenA.I. Sandbox and the broader digital banking landscape in Hong Kong. We previously covered how Ant International and Bettr Join Hong Kong GenA.I. Sandbox++ for AI Liquidity Management, alongside similar moves from insurers like AXA and Manulife. Regarding the bank itself, we reported on its efforts to support the SME sector when Ping An Digital Bank Launches Collateral-Free Purchase Order Financing for E-commerce SMEs. This ongoing coverage highlights a regional trend where digital banks are moving beyond basic services to tackle complex institutional and security challenges.

What this means

The success of this "A.I. vs A.I." use case signals a shift in the fintech security paradigm. As generative A.I. makes high-quality deepfakes accessible to bad actors, traditional biometric verification is under immense pressure. Ping An’s 96% accuracy rate sets a high benchmark, but it also underscores the reality that no system is yet 100% foolproof. For the wider industry, this announcement places pressure on legacy institutions to accelerate their own A.I. adoption or risk becoming the path of least resistance for fraudsters. It raises a critical question: as defensive A.I. improves, will attackers simply pivot to more sophisticated, low-volume social engineering that bypasses technical detection entirely?