• The Challenge: Deepfakes undermine both the accuracy and fairness of AI face detection due to variations in human features.
• The Technology: Ant International’s AI model uses a novel ‘Mixture of Experts’-based Adversarial Debiasing approach to achieve superior fairness and accuracy across diverse demographics.
• The Impact: This advancement strengthens payment security, offering robust protection against deepfake fraud and enhancing eKYC compliance globally.

Ant International, a leading global digital payment, digitisation, and financial technology provider, has won first place at the NeurIPS Competition of Fairness in AI Face Detection.

AI Fairness enhances Financial Security

Studies by the National Institute of Standards and Technology (NIST) show that many commercial facial recognition algorithms have significantly higher error rates for women and people of colour largely due to underrepresentation in datasets AI algorithms are trained on. Biased algorithms lead to unfair denials of service and create security vulnerabilities.

The competition at the Conference on Neural Information Processing Systems, among the world’s most prestigious AI conferences, challenged participants to develop AI models that not only achieve high-utility performance but also demonstrate fairness across demographic subgroups such as gender, age, and skin tone. Ant International emerged first among over 2,100 submissions from 162 teams across the world at the competition. These teams were tasked to detect 1.2 million AI-generated face pictures of humans from different demographic groups.

By combining a Mixture of Experts (MoE) architecture with an internal “bias-detection” mechanism, Ant International’s winning AI model trains competing neural sub-networks – one focused on spotting deepfakes, the other actively challenging it to ignore sensitive traits. This dynamic ensures the system learns to detect only genuine signs of manipulation, not demographic patterns. Trained on globally representative data including real-world payment fraud scenarios, the model delivers robust and equitable performance at scale.

“A biased AI is an insecure AI,” said Dr. Tianyi Zhang, General Manager of Risk Management and Cybersecurity at Ant International. “Our model’s fairness prevents exploitation from deepfakes and ensures reliable identity verification for all users, supporting our mission to deliver secure and inclusive financial services worldwide.”

Applications in Financial Services

Ant International is integrating the NeurIPS-winning technology directly into its payment and financial services to counter deepfake threats and ensure consistent customer experience globally.

• Account protection against deepfake attacks: Ant International’s AI-powered anti-deepfake measures have achieved a detection rate exceeding 99.8%. By prioritising fairness, this ensures robust and consistent protection against sophisticated threats like account takeovers for all demographics across 200 markets worldwide.

• Enhanced compliance for eKYC processes: The company, as well as its financial institution and merchant partners, can leverage the technology to meet stringent global Electronic Know Your Customer (eKYC) standards in customer onboarding across different markets, without introducing algorithmic bias.

• Advancing financial inclusion in emerging markets: The development of equitable and accurate AI helps ensure that Ant International can extend services to a broader, more diverse market globally, fostering greater financial inclusion.

Full-cycle security strategy to protect financial transactions

Ant International alongside its partners serve over 150 million merchants and 1.8 billion user accounts worldwide. These services include global wallet gateway Alipay+, merchant payment services Antom, inclusive finance and credit tech services Bettr and global account services WorldFirst.

Ant International has positioned AI security at the core of its operations. It has introduced the AI SHIELD, as a comprehensive, full-lifecycle foundational framework for risk management. Built on the AI Security Docker, this solution reduces risk of AI service vulnerabilities such as unauthorised access and data leakage by 90% through advanced assessment methods and running safeguards.

Risk-management solutions secured by AI SHIELD provide full-lifecycle protection of transactions at Ant International. These solutions protect businesses and users from deepfake attacks, frauds and scams. For instance, account protection solution Alipay+ EasySafePay 360 can reduce account takeover incidence for digital wallet payments by 90%.

Source: BUSINESS WIRE